fix: fix test error

### [Version 1.123.2](https://github.com/lobehub/lobe-chat/compare/v1.123.1...v1.123.2)
<sup>Released on **2025-09-05**</sup>

#### 🐛 Bug Fixes

- **misc**: Not use branch topic when this topic is not save.

<br/>

<details>
<summary><kbd>Improvements and Fixes</kbd></summary>

#### What's fixed

* **misc**: Not use branch topic when this topic is not save, closes [#9083](https://github.com/lobehub/lobe-chat/issues/9083) ([f534d19](https://github.com/lobehub/lobe-chat/commit/f534d19))

</details>

<div align="right">

[![](https://img.shields.io/badge/-BACK_TO_TOP-151515?style=flat-square)](#readme-top)

</div>

.cursor/rules/add-provider-doc.mdc

@@ -0,0 +1,183 @@
+---
+description: Complete guide for adding a new AI provider documentation to LobeChat
+alwaysApply: false
+---
+
+# Adding New AI Provider Documentation
+
+This document provides a step-by-step guide for adding documentation for a new AI provider to LobeChat, based on the complete workflow used for adding providers like BFL (Black Forest Labs) and FAL.
+
+## Overview
+
+Adding a new provider requires creating both user-facing documentation and technical configuration files. The process involves:
+
+1. Creating usage documentation (EN + CN)
+2. Adding environment variable documentation (EN + CN)
+3. Updating Docker configuration files
+4. Updating .env.example file
+5. Preparing image resources
+
+## Step 1: Create Provider Usage Documentation
+
+Create user-facing documentation that explains how to use the new provider.
+
+### Required Files
+
+Create both English and Chinese versions:
+- `docs/usage/providers/{provider-name}.mdx` (English)
+- `docs/usage/providers/{provider-name}.zh-CN.mdx` (Chinese)
+
+### Documentation Structure
+
+Follow the structure and format used in existing provider documentation. For reference, see:
+- `docs/usage/providers/fal.mdx` (English template)
+- `docs/usage/providers/fal.zh-CN.mdx` (Chinese template)
+
+### Key Requirements
+
+- **Images**: Prepare 5-6 screenshots showing the process
+- **Cover Image**: Create or obtain a cover image for the provider
+- **Accurate URLs**: Use real registration and dashboard URLs
+- **Service Type**: Specify whether it's for image generation, text generation, etc.
+- **Pricing Warning**: Include pricing information callout
+
+### Important Notes
+
+- **🔒 API Key Security**: Never include real API keys in documentation. Always use placeholder format (e.g., `bfl-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx`)
+- **🖼️ Image Hosting**: Use LobeHub's CDN for all images: `hub-apac-1.lobeobjects.space`
+
+## Step 2: Update Environment Variables Documentation
+
+Add the new provider's environment variables to the self-hosting documentation.
+
+### Files to Update
+
+- `docs/self-hosting/environment-variables/model-provider.mdx` (English)
+- `docs/self-hosting/environment-variables/model-provider.zh-CN.mdx` (Chinese)
+
+### Content to Add
+
+Add two sections for each provider:
+
+```markdown
+### `{PROVIDER}_API_KEY`
+
+- Type: Required
+- Description: This is the API key you applied for in the {Provider Name} service.
+- Default: -
+- Example: `{api-key-format-example}`
+
+### `{PROVIDER}_MODEL_LIST`
+
+- Type: Optional
+- Description: Used to control the {Provider Name} model list. Use `+` to add a model, `-` to hide a model, and `model_name=display_name` to customize the display name of a model. Separate multiple entries with commas. The definition syntax follows the same rules as other providers' model lists.
+- Default: `-`
+- Example: `-all,+{model-id-1},+{model-id-2}={display-name}`
+
+The above example disables all models first, then enables `{model-id-1}` and `{model-id-2}` (displayed as `{display-name}`).
+
+[model-list]: /docs/self-hosting/advanced/model-list
+```
+
+### Important Notes
+
+- **API Key Format**: Use proper UUID format for examples (e.g., `12345678-1234-1234-1234-123456789abc`)
+- **Real Model IDs**: Use actual model IDs from the codebase, not placeholders
+- **Consistent Naming**: Follow the pattern `{PROVIDER}_API_KEY` and `{PROVIDER}_MODEL_LIST`
+
+## Step 3: Update Docker Configuration Files
+
+Add environment variables to all Docker configuration files to ensure the provider works in containerized deployments.
+
+### Files to Update
+
+All Dockerfile variants must be updated:
+- `Dockerfile`
+- `Dockerfile.database`
+- `Dockerfile.pglite`
+
+### Changes Required
+
+Add the new provider's environment variables at the **end** of the ENV section, just before the final line:
+
+```dockerfile
+# Previous providers...
+    # 302.AI
+    AI302_API_KEY="" AI302_MODEL_LIST="" \
+    # {New Provider 1}
+    {PROVIDER1}_API_KEY="" {PROVIDER1}_MODEL_LIST="" \
+    # {New Provider 2}
+    {PROVIDER2}_API_KEY="" {PROVIDER2}_MODEL_LIST=""
+```
+
+### Important Rules
+
+- **Position**: Add new providers at the **end** of the list
+- **Ordering**: When adding multiple providers, use alphabetical order (e.g., FAL before BFL)
+- **Consistency**: Maintain identical ordering across all Dockerfile variants
+- **Format**: Follow the pattern `{PROVIDER}_API_KEY="" {PROVIDER}_MODEL_LIST="" \`
+
+## Step 4: Update .env.example File
+
+Add example configuration entries to help users understand how to configure the provider locally.
+
+### File to Update
+
+- `.env.example`
+
+### Content to Add
+
+Add new sections before the "Market Service" section:
+
+```bash
+### {Provider Name} ###
+
+# {PROVIDER}_API_KEY={provider-prefix}-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
+```
+
+### Format Guidelines
+
+- **Section Header**: Use `### {Provider Name} ###` format
+- **Commented Example**: Use `#` to comment out the example
+- **Key Format**: Use appropriate prefix for the provider (e.g., `bfl-`, `fal-`, `sk-`)
+- **Position**: Add before the Market Service section
+- **Spacing**: Maintain consistent spacing with existing entries
+
+## Step 5: Image Resources
+
+Prepare all necessary image resources for the documentation.
+
+### Required Images
+
+1. **Cover Image**: Provider logo or branded image
+2. **API Dashboard Screenshots**: 3-4 screenshots showing API key creation process
+3. **LobeChat Configuration Screenshots**: 2-3 screenshots showing provider setup in LobeChat
+
+### Image Guidelines
+
+- **Quality**: Use high-resolution screenshots
+- **Consistency**: Maintain consistent styling across all screenshots
+- **Privacy**: Remove or blur any sensitive information
+- **Format**: Use PNG format for screenshots
+- **Hosting**: Use LobeHub's CDN (`hub-apac-1.lobeobjects.space`) for all images
+
+## Checklist
+
+Before submitting your provider documentation:
+
+- [ ] Created both English and Chinese usage documentation
+- [ ] Added environment variable documentation (EN + CN)
+- [ ] Updated all 3 Dockerfile variants with consistent ordering
+- [ ] Updated .env.example with proper key format
+- [ ] Prepared all required screenshots and images
+- [ ] Used actual model IDs from the codebase
+- [ ] Verified no real API keys are included in documentation
+- [ ] Used LobeHub CDN for all image hosting
+- [ ] Tested the documentation for clarity and accuracy
+
+## Reference
+
+This guide was created based on the implementation of BFL (Black Forest Labs) provider documentation. For a complete example, refer to:
+- Commits: `d2da03e1a` (documentation) and `6a2e95868` (environment variables)
+- Files: `docs/usage/providers/bfl.mdx`, `docs/usage/providers/bfl.zh-CN.mdx`
+- PR: Current branch `tj/feat/bfl-docs`

.cursor/rules/debug.mdc

@@ -1,193 +0,0 @@
----
-description: Debug 调试指南
-globs: 
-alwaysApply: false
----
-# Debug 调试指南
-
-## 💡 调试流程概览
-
-当遇到问题时，请按照以下优先级进行处理：
-
-1. **快速判断** - 对于熟悉的错误，直接提供解决方案
-2. **信息收集** - 使用工具搜索相关代码和配置
-3. **网络搜索** - 查找现有解决方案
-4. **定位调试** - 添加日志进行问题定位
-5. **临时方案** - 如果找不到根本解决方案，提供临时解决方案
-6. **解决实施** - 提供可维护的最终解决方案
-
-## 🔍 错误信息分析
-
-### 错误来源识别
-
-错误信息可能来自：
-
-- **Terminal 输出** - 构建、运行时错误
-- **浏览器控制台** - 前端 JavaScript 错误
-- **开发工具** - ESLint、TypeScript、测试框架等
-- **服务器日志** - API、数据库连接等后端错误
-- **截图或文本** - 用户直接提供的错误信息
-
-## 🛠️ 信息收集工具
-
-### 代码搜索工具
-
-使用以下工具收集相关信息，并根据场景选择最合适的工具：
-
-- **`codebase_search` (语义搜索)**
-    - **何时使用**: 当你不确定具体的代码实现，想要寻找相关概念、功能或逻辑时。
-    - **示例**: `查询"文件上传"功能的实现`
-- **`grep_search` (精确/正则搜索)**
-    - **何时使用**: 当你知道要查找的确切字符串、函数名、变量名或一个特定的模式时。
-    - **示例**: `查找所有使用了 'useState' 的地方`
-- **`file_search` (文件搜索)**
-    - **何时使用**: 当你知道文件名的一部分，需要快速定位文件时。
-    - **示例**: `查找 'Button.tsx' 组件`
-- **`read_file` (内容读取)**
-    - **何时使用**: 在定位到具体文件后，用于查看其完整内容和上下文。
-- **`web_search` (网络搜索)**
-    - **何时使用**: 当错误信息可能与第三方库、API 或常见问题相关时，用于获取外部信息。
-
-### 环境与依赖检查
-
-- **检查 `package.json`**: 查看 `scripts` 了解项目如何运行、构建和测试。查看 `dependencies` 和 `devDependencies` 确认库版本，版本冲突有时是问题的根源。
-- **运行测试**: 使用 `ni vitest` 运行单元测试和集成测试，这可以快速定位功能回归或组件错误。
-
-### 项目特定搜索目标
-
-针对 lobe-chat 项目，重点关注：
-
-- **配置文件**: [package.json](mdc:package.json), [next.config.mjs](mdc:next.config.mjs)
-- **核心功能**: `src/features/` 下的相关模块
-- **状态管理**: `src/store/` 下的 Zustand stores
-- **数据库**: `src/database/` 和 `src/migrations/`
-- **类型定义**: `src/types/` 下的类型文件
-- **服务层**: `src/services/` 下的 API 服务
-- **启动流程**: [apps/desktop/src/main/core/App.ts](mdc:apps/desktop/src/main/core/App.ts) - 了解应用启动流程
-
-## 🌐 网络搜索策略
-
-### 搜索顺序优先级
-
-1. **和问题相关的项目的 github issue**
-
-2. **技术社区**
-    - Stack Overflow
-    - GitHub Discussions
-    - Reddit
-
-3. **官方文档**
-    - 使用 `mcp_context7_resolve-library-id` 和 `mcp_context7_get-library-docs` 工具
-    - 查阅官方文档网站
-
-### 搜索关键词策略
-
-- **错误信息**: 完整的错误消息
-- **技术栈**: "Next.js 15" + "error message"
-- **上下文**: 添加功能相关的关键词
-
-## 🔧 问题定位与结构化思考
-
-如果问题比较复杂，我们要按照先定位问题，再解决问题的大方向进行。
-
-### 结构化思考工具
-
-对于复杂或多步骤的调试任务，使用 `mcp_sequential-thinking_sequentialthinking` 工具来结构化思考过程。这有助于：
-
-- **分解问题**: 将大问题拆解成可管理的小步骤。
-- **清晰追踪**: 记录每一步的发现和决策，避免遗漏。
-- **自我修正**: 在过程中评估和调整调试路径。
-
-### 日志调试
-
-在问题产生的路径上添加日志，可以简单使用 `console.log` 或者参考 [debug-usage.mdc](mdc:.cursor/rules/debug-usage.mdc) 使用 `debug` 模块。添加完日志后，请求我运行相关的代码并提供关键输出和错误信息。
-
-### 引导式交互调试
-
-虽然我无法直接操作浏览器开发者工具，但我可以引导你进行交互式调试：
-
-1.  **设置断点**: 我会告诉你可以在哪些关键代码行设置断点。
-2.  **检查变量**: 我会请你在断点处检查特定变量的值或 `props`/`state`。
-3.  **分析调用栈**: 我会请你提供调用栈信息，以帮助理解代码执行流程。
-
-## 💡 临时解决方案策略
-
-当无法找到根本解决方案时，提供临时解决方案：
-
-### 临时方案准则
-
-- **快速修复** - 优先让功能可用
-- **最小修改** - 减少对现有代码的影响
-- **清晰标记** - 明确标注这是临时方案
-- **后续计划** - 说明后续如何找到更好的解决方案
-
-### 临时方案模板
-
-```markdown
-## 临时解决方案 ⚠️
-
-**问题**: [简要描述问题]
-
-**临时修复**:
-[具体的临时修复步骤]
-
-**风险说明**:
-
-- [可能的副作用或限制]
-- [需要注意的事项]
-
-**后续计划**:
-
-- [ ] 深入调研根本原因
-- [ ] 寻找更优雅的解决方案
-- [ ] 监控是否有其他影响
-```
-
-## ✅ 解决方案准则
-
-### 方案质量标准
-
-提供的解决方案应该：
-
-- **✅ 低侵入性** - 最小化对现有代码的修改
-- **✅ 可维护性** - 易于理解和后续维护
-- **✅ 类型安全** - 符合 TypeScript 规范
-- **✅ 最佳实践** - 遵循项目的编码规范
-- **✅ 测试友好** - 便于编写和运行测试
-- **❌ 避免长期 Hack** - 临时方案可以 hack，但要明确标注
-
-### 解决方案模板
-
-```markdown
-## 问题原因
-
-[简要说明问题产生的根本原因]
-
-## 解决方案
-
-[详细的解决步骤]
-
-## 代码修改
-
-[具体的代码变更]
-
-## 验证方法
-
-[如何验证问题已解决]
-
-## 预防措施
-
-[如何避免类似问题再次发生]
-```
-
-## 🔄 迭代调试流程
-
-如果初次解决方案无效：
-
-1.  **重新收集信息** - 基于新的错误信息搜索
-2.  **深入代码分析** - 查看更多相关代码文件
-3.  **运行相关测试** - 编写或运行一个失败的测试来稳定复现问题。
-4.  **扩大搜索范围** - 搜索更广泛的相关问题
-5.  **请求更多日志** - 添加更详细的调试信息
-6.  **提供临时方案** - 如果根本解决方案复杂，先提供临时修复
-7.  **分解问题** - 将复杂问题拆解为更小的子问题

.cursor/rules/project-introduce.mdc

@@ -10,7 +10,7 @@ Emoji logo: 🤯
 
 ## Project Technologies Stack
 
-read [package.json](mdc:package.json) to know all npm packages you can use. read [folder-structure.mdx](mdc:docs/development/basic/folder-structure.mdx) to learn project structure.
+read [package.json](mdc:package.json) to know all npm packages you can use.
 
 The project uses the following technologies:
 
@@ -42,17 +42,3 @@ The project uses the following technologies:
 - Cursor AI for code editing and AI coding assistance
 
 Note: All tools and libraries used are the latest versions. The application only needs to be compatible with the latest browsers;
-
-## Often used npm scripts and commands
-
-```bash
-# !: don't  any build script to check weather code can work after modify
-# type check
-bun run type-check
-
-# install dependencies
-pnpm install
-
-# run tests
-npx vitest run --config vitest.config.ts '[file-path-pattern]'
-```

.cursor/rules/project-structure.mdc

@@ -0,0 +1,239 @@
+---
+description: Project directory structure overview
+alwaysApply: false
+---
+
+# LobeChat Project Structure
+
+## Directory Structure
+
+note: some files are not shown for simplicity.
+
+```plaintext
+lobe-chat/
+├── apps/                           # Applications directory
+│   └── desktop/                    # Electron desktop application
+│       ├── src/                    # Desktop app source code
+│       └── resources/              # Desktop app resources
+├── docs/                           # Project documentation
+│   ├── development/                # Development docs
+│   ├── self-hosting/               # Self-hosting docs
+│   └── usage/                      # Usage guides
+├── locales/                        # Internationalization files (multiple locales)
+│   ├── en-US/                      # English (example)
+│   └── zh-CN/                      # Simplified Chinese (example)
+├── packages/                       # Monorepo packages directory
+│   ├── const/                      # Constants definition package
+│   ├── database/                   # Database related package
+│   ├── electron-client-ipc/        # Electron renderer ↔ main IPC client
+│   ├── electron-server-ipc/        # Electron main process IPC server
+│   ├── model-bank/                 # Built-in model presets/catalog exports
+│   ├── model-runtime/              # AI model runtime package
+│   ├── types/                      # TypeScript type definitions
+│   ├── utils/                      # Utility functions package
+│   ├── file-loaders/               # File processing packages
+│   ├── prompts/                    # AI prompt management
+│   └── web-crawler/                # Web crawling functionality
+├── public/                         # Static assets
+│   ├── icons/                      # Application icons
+│   ├── images/                     # Image resources
+│   └── screenshots/                # Application screenshots
+├── scripts/                        # Build and tool scripts
+├── src/                            # Main application source code (see below)
+├── .cursor/                        # Cursor AI configuration
+├── docker-compose/                 # Docker configuration
+├── package.json                    # Project dependencies
+├── pnpm-workspace.yaml            # pnpm monorepo configuration
+├── next.config.ts                  # Next.js configuration
+├── drizzle.config.ts              # Drizzle ORM configuration
+└── tsconfig.json                   # TypeScript configuration
+```
+
+## Core Source Directory (`src/`)
+
+```plaintext
+src/
+├── app/                            # Next.js App Router routes
+│   ├── (backend)/                  # Backend API routes
+│   │   ├── api/                    # REST API endpoints
+│   │   │   ├── auth/               # Authentication endpoints
+│   │   │   └── webhooks/           # Webhook handlers for various auth providers
+│   │   ├── middleware/             # Request middleware
+│   │   ├── oidc/                   # OpenID Connect endpoints
+│   │   ├── trpc/                   # tRPC API routes
+│   │   │   ├── async/              # Async tRPC endpoints
+│   │   │   ├── desktop/            # Desktop runtime endpoints
+│   │   │   ├── edge/               # Edge runtime endpoints
+│   │   │   ├── lambda/             # Lambda runtime endpoints
+│   │   │   └── tools/              # Tools-specific endpoints
+│   │   └── webapi/                 # Web API endpoints
+│   │       ├── chat/               # Chat-related APIs for various providers
+│   │       ├── models/             # Model management APIs
+│   │       ├── plugin/             # Plugin system APIs
+│   │       ├── stt/                # Speech-to-text APIs
+│   │       ├── text-to-image/      # Image generation APIs
+│   │       └── tts/                # Text-to-speech APIs
+│   ├── [variants]/                 # Page route variants
+│   │   ├── (main)/                 # Main application routes
+│   │   │   ├── chat/               # Chat interface and workspace
+│   │   │   ├── discover/           # Discover page (assistants, models, providers)
+│   │   │   ├── files/              # File management interface
+│   │   │   ├── image/              # Image generation interface
+│   │   │   ├── profile/            # User profile and stats
+│   │   │   ├── repos/              # Knowledge base repositories
+│   │   │   └── settings/           # Application settings
+│   │   └── @modal/                 # Modal routes
+│   └── manifest.ts                 # PWA configuration
+├── components/                     # Global shared components
+│   ├── Analytics/                  # Analytics tracking components
+│   ├── Error/                      # Error handling components
+│   └── Loading/                    # Loading state components
+├── config/                         # Application configuration
+│   ├── featureFlags/               # Feature flags & experiments
+│   └── modelProviders/             # Model provider configurations
+├── features/                       # Feature components (UI Layer)
+│   ├── AgentSetting/               # Agent configuration and management
+│   ├── ChatInput/                  # Chat input with file upload and tools
+│   ├── Conversation/               # Message display and interaction
+│   ├── FileManager/                # File upload and knowledge base
+│   └── PluginStore/                # Plugin marketplace and management
+├── hooks/                          # Custom React hooks
+├── layout/                         # Global layout components
+│   ├── AuthProvider/               # Authentication provider
+│   └── GlobalProvider/             # Global state provider
+├── libs/                           # External library integrations
+│   ├── analytics/                  # Analytics services integration
+│   ├── next-auth/                  # NextAuth.js configuration
+│   └── oidc-provider/              # OIDC provider implementation
+├── locales/                        # Internationalization resources
+│   └── default/                    # Default language definitions
+├── migrations/                     # Client-side data migrations
+├── server/                         # Server-side code
+│   ├── modules/                    # Server modules
+│   ├── routers/                    # tRPC routers
+│   └── services/                   # Server services
+├── services/                       # Service layer (per-domain, client/server split)
+│   ├── user/                       # User services
+│   │   ├── client.ts               # Client DB (PGLite) implementation
+│   │   └── server.ts               # Server DB implementation (via tRPC)
+│   ├── aiModel/                    # AI model services
+│   ├── session/                    # Session services
+│   └── message/                    # Message services
+├── store/                          # Zustand state management
+│   ├── agent/                      # Agent state
+│   ├── chat/                       # Chat state
+│   └── user/                       # User state
+├── styles/                         # Global styles
+├── tools/                          # Built-in tool system
+│   ├── artifacts/                  # Code artifacts and preview
+│   └── web-browsing/               # Web search and browsing
+├── types/                          # TypeScript type definitions
+└── utils/                          # Utility functions
+    ├── client/                     # Client-side utilities
+    └── server/                     # Server-side utilities
+```
+
+## Key Monorepo Packages
+
+```plaintext
+packages/
+├── const/                          # Global constants and configurations
+├── database/                       # Database schemas and models
+│   ├── src/models/                 # Data models (CRUD operations)
+│   ├── src/schemas/                # Drizzle database schemas
+│   ├── src/repositories/           # Complex query layer
+│   └── migrations/                 # Database migration files
+├── model-runtime/                  # AI model runtime
+│   └── src/
+│       ├── openai/                 # OpenAI provider integration
+│       ├── anthropic/              # Anthropic provider integration
+│       ├── google/                 # Google AI provider integration
+│       ├── ollama/                 # Ollama local model integration
+│       ├── types/                  # Runtime type definitions
+│       └── utils/                  # Runtime utilities
+├── types/                          # Shared TypeScript type definitions
+│   └── src/
+│       ├── agent/                  # Agent-related types
+│       ├── message/                # Message and chat types
+│       ├── user/                   # User and session types
+│       └── tool/                   # Tool and plugin types
+├── utils/                          # Shared utility functions
+│   └── src/
+│       ├── client/                 # Client-side utilities
+│       ├── server/                 # Server-side utilities
+│       ├── fetch/                  # HTTP request utilities
+│       └── tokenizer/              # Token counting utilities
+├── file-loaders/                   # File loaders (PDF, DOCX, etc.)
+├── prompts/                        # AI prompt management
+└── web-crawler/                    # Web crawling functionality
+```
+
+## Architecture Map
+
+- Presentation: `src/features`, `src/components`, `src/layout` — UI composition, global providers
+- State: `src/store` — Zustand slices, selectors, middleware
+- Client Services: `src/services/<domain>/{client|server}.ts` — client: PGLite; server: tRPC bridge
+- API Routers: `src/app/(backend)/webapi` (REST), `src/app/(backend)/trpc/{edge|lambda|async|desktop|tools}`; Lambda router triggers Async router for long-running tasks (e.g., image)
+- Server Services: `src/server/services` (business logic), `src/server/modules` (infra adapters)
+- Data Access: `packages/database/src/{schemas,models,repositories}` — Schema (Drizzle), Model (CRUD), Repository (complex queries)
+- Integrations: `src/libs` — analytics, auth, trpc, logging, runtime helpers
+
+## Data Flow Architecture
+
+### Unified Flow Pattern
+
+```
+UI Layer → State Management → Client Service → [Environment Branch] → Database
+   ↓              ↓               ↓                    ↓                ↓
+ React         Zustand     Environment         Local/Remote        PGLite/
+Components      Store      Adaptation            Routing          PostgreSQL
+```
+
+### Environment-Specific Routing
+
+| Mode            | UI       | Service Route          | Database            |
+| --------------- | -------- | ---------------------- | ------------------- |
+| **Browser/PWA** | React    | Direct Model Access    | PGLite (Local)      |
+| **Server**      | React    | tRPC → Server Services | PostgreSQL (Remote) |
+| **Desktop**     | Electron | tRPC → Local Node.js   | PGLite/PostgreSQL\* |
+
+_\*Depends on cloud sync configuration_
+
+### Key Characteristics
+
+- **Type Safety**: End-to-end type safety via tRPC and Drizzle ORM
+- **Local/Remote Dual Mode**: PGLite enables user data ownership and local control
+
+## Quick Map
+
+- App Routes: `src/app` — UI routes (App Router) and backend routes under `(backend)`
+- Web API: `src/app/(backend)/webapi` — REST-like endpoints
+- tRPC Routers: `src/server/routers` — typed RPC endpoints by runtime
+- Client Services: `src/services` — environment-adaptive client-side business logic
+- Server Services: `src/server/services` — platform-agnostic business logic
+- Database: `packages/database` — schemas/models/repositories/migrations
+- State: `src/store` — Zustand stores and slices
+- Integrations: `src/libs` — analytics/auth/trpc/logging/runtime helpers
+- Tools: `src/tools` — built-in tool system
+
+## Common Tasks
+
+- Add Web API route: `src/app/(backend)/webapi/<module>/route.ts`
+- Add tRPC endpoint: `src/server/routers/{edge|lambda|desktop}/...`
+- Add client/server service: `src/services/<domain>/{client|server}.ts` (client: PGLite; server: tRPC)
+- Add server service: `src/server/services/<domain>`
+- Add a new model/provider: `src/config/modelProviders/<provider>.ts` + `packages/model-bank/src/aiModels/<provider>.ts` + `packages/model-runtime/src/<provider>/index.ts`
+- Add DB schema/model/repository: `packages/database/src/{schemas|models|repositories}`
+- Add Zustand slice: `src/store/<domain>/slices`
+
+## Env Modes
+
+- `NEXT_PUBLIC_CLIENT_DB`: selects client DB mode (e.g., `pglite`) vs server-backed
+- `NEXT_PUBLIC_IS_DESKTOP_APP`: enables desktop-specific routes and behavior
+- `NEXT_PUBLIC_SERVICE_MODE`: controls service routing preference (client/server)
+
+## Boundaries
+
+- Keep client logic in `src/services`; server-only logic stays in `src/server/services`
+- Don’t mix Web API (`webapi/`) with tRPC (`src/server/routers/`)
+- Place business UI under `src/features`, global reusable UI under `src/components`

.cursor/rules/testing-guide/db-model-test.mdc

@@ -5,6 +5,8 @@ alwaysApply: false
 
 ## 🗃️ 数据库 Model 测试指南
 
+测试 `packages/database` 下的数据库 Model 层。
+
 ### 测试环境选择 💡
 
 数据库 Model 层通过环境变量控制数据库类型，在两种测试环境下有不同的数据库后端：客户端环境 (PGLite) 和 服务端环境 (PostgreSQL)
@@ -17,10 +19,10 @@ alwaysApply: false
 
 ```bash
 # 1. 先在客户端环境测试（快速验证）
-npx vitest run --config vitest.config.ts src/database/models/__tests__/myModel.test.ts
+cd packages/database && TEST_SERVER_DB=0 bunx vitest run --silent='passed-only' src/database/models/__tests__/myModel.test.ts
 
-# 2. 再在服务端环境测试（兼容性验证）
-npx vitest run --config vitest.config.server.ts src/database/models/__tests__/myModel.test.ts
+# 2. 再在服务端环境测试（兼容性验证）, 需要设置环境变量 `TEST_SERVER_DB=1`
+cd packages/database && TEST_SERVER_DB=1 bunx vitest run --silent='passed-only' src/database/models/__tests__/myModel.test.ts #
 ```
 
 ### 创建新 Model 测试的最佳实践 📋

.cursor/rules/testing-guide/testing-guide.mdc

@@ -5,11 +5,11 @@ alwaysApply: false
 
 # 测试指南 - LobeChat Testing Guide
 
-## 🧪 测试环境概览
+## 测试环境概览
 
 LobeChat 项目使用 Vitest 测试库，配置了两种不同的测试环境：
 
-### 客户端测试环境 (DOM Environment)
+### 客户端数据库测试环境 (DOM Environment)
 
 - **配置文件**: [vitest.config.ts](mdc:vitest.config.ts)
 - **环境**: Happy DOM (浏览器环境模拟)
@@ -17,60 +17,71 @@ LobeChat 项目使用 Vitest 测试库，配置了两种不同的测试环境：
 - **用途**: 测试前端组件、客户端逻辑、React 组件等
 - **设置文件**: [tests/setup.ts](mdc:tests/setup.ts)
 
-### 服务端测试环境 (Node Environment)
+### 服务端数据库测试环境 (Node Environment)
 
-- **配置文件**: [vitest.config.server.ts](mdc:vitest.config.server.ts)
+目前只有 `packages/database` 下的测试可以通过配置 `TEST_SERVER_DB=1` 环境变量来使用服务端数据库测试
+
+- **配置文件**: [packages/database/vitest.config.mts](mdc:packages/database/vitest.config.mts) 并且设置环境变量 `TEST_SERVER_DB=1`
 - **环境**: Node.js
 - **数据库**: 真实的 PostgreSQL 数据库
 - **并发限制**: 单线程运行 (`singleFork: true`)
 - **用途**: 测试数据库模型、服务端逻辑、API 端点等
-- **设置文件**: [tests/setup-db.ts](mdc:tests/setup-db.ts)
+- **设置文件**: [packages/database/tests/setup-db.ts](mdc:packages/database/tests/setup-db.ts)
 
-## 🚀 测试运行命令
+## 测试运行命令
 
-**🚨 性能警告**: 项目包含 3000+ 测试用例，完整运行需要约 10 分钟。务必使用文件过滤或测试名称过滤。
+** 性能警告**: 项目包含 3000+ 测试用例，完整运行需要约 10 分钟。务必使用文件过滤或测试名称过滤。
 
-### ✅ 正确的命令格式
+### 正确的命令格式
 
 ```bash
 # 运行所有客户端/服务端测试
-npx vitest run --config vitest.config.ts        # 客户端测试
-npx vitest run --config vitest.config.server.ts # 服务端测试
+bunx vitest run --silent='passed-only'                                          # 客户端测试
+cd packages/database && TEST_SERVER_DB=1 bunx vitest run --silent='passed-only' # 服务端测试
 
 # 运行特定测试文件 (支持模糊匹配)
-npx vitest run --config vitest.config.ts user.test.ts
+bunx vitest run --silent='passed-only' user.test.ts
 
 # 运行特定测试用例名称 (使用 -t 参数)
-npx vitest run --config vitest.config.ts -t "test case name"
+bunx vitest run --silent='passed-only' -t "test case name"
 
 # 组合使用文件和测试名称过滤
-npx vitest run --config vitest.config.ts filename.test.ts -t "specific test"
+bunx vitest run --silent='passed-only' filename.test.ts -t "specific test"
 
 # 生成覆盖率报告 (使用 --coverage 参数)
-npx vitest run --config vitest.config.ts --coverage
+bunx vitest run --silent='passed-only' --coverage
 ```
 
-### ❌ 避免的命令格式
+### 避免的命令格式
 
 ```bash
-# ❌ 这些命令会运行所有 3000+ 测试用例，耗时约 10 分钟！
+#  这些命令会运行所有 3000+ 测试用例，耗时约 10 分钟！
 npm test
 npm test some-file.test.ts
 
-# ❌ 不要使用裸 vitest (会进入 watch 模式)
+#  不要使用裸 vitest (会进入 watch 模式)
 vitest test-file.test.ts
 ```
 
-## 🔧 测试修复原则
+## 测试修复原则
 
-### 核心原则 ⚠️
+### 核心原则
 
-1. **充分阅读测试代码**: 在修复测试之前，必须完整理解测试的意图和实现
-2. **测试优先修复**: 如果是测试本身写错了，修改测试而不是实现代码
-3. **专注单一问题**: 只修复指定的测试，不要添加额外测试或功能
-4. **不自作主张**: 不要因为发现其他问题就直接修改，先提出再讨论
+1. **收集足够的上下文**  
+   在修复测试之前，务必做到：
+   - 完整理解测试的意图和实现
+   - 强烈建议阅读当前的 git diff 和 PR diff
 
-### 测试协作最佳实践 🤝
+2. **测试优先修复**  
+   如果是测试本身写错了，应优先修改测试，而不是实现代码。
+
+3. **专注单一问题**  
+   只修复指定的测试，不要顺带添加额外测试。
+
+4. **不自作主张**  
+   发现其他问题时，不要直接修改，需先提出并讨论。
+
+### 测试协作最佳实践
 
 基于实际开发经验总结的重要协作原则：
 
@@ -84,10 +95,10 @@ vitest test-file.test.ts
 - **避免陷阱**: 不要陷入"不断尝试相同或类似方法"的循环
 
 ```typescript
-// ❌ 错误做法：连续失败后继续盲目尝试
+//  错误做法：连续失败后继续盲目尝试
 // 第3次、第4次仍在用相似的方法修复同一个问题
 
-// ✅ 正确做法：失败1-2次后总结问题
+//  正确做法：失败1-2次后总结问题
 /*
 问题总结：
 1. 尝试过的方法：修改 mock 数据结构
@@ -106,7 +117,7 @@ vitest test-file.test.ts
 - **保持稳定性**: 测试名称应该在代码重构后仍然有意义
 
 ```typescript
-// ❌ 错误的测试命名
+//  错误的测试命名
 describe('User component coverage', () => {
   it('covers line 45-50 in getUserData', () => {
     // 为了覆盖第45-50行而写的测试
@@ -117,7 +128,7 @@ describe('User component coverage', () => {
   });
 });
 
-// ✅ 正确的测试命名
+//  正确的测试命名
 describe('<UserAvatar />', () => {
   it('should render fallback icon when image url is not provided', () => {
     // 测试具体的业务场景，自然会覆盖相关代码分支
@@ -131,8 +142,8 @@ describe('<UserAvatar />', () => {
 
 **覆盖率提升的正确思路**:
 
-- ✅ 通过设计各种业务场景（正常流程、边缘情况、错误处理）来自然提升覆盖率
-- ❌ 不要为了达到覆盖率数字而写测试，更不要在测试中注释"为了覆盖 xxx 行"
+- 通过设计各种业务场景（正常流程、边缘情况、错误处理）来自然提升覆盖率
+- 不要为了达到覆盖率数字而写测试，更不要在测试中注释"为了覆盖 xxx 行"
 
 #### 3. 测试组织结构
 
@@ -143,7 +154,7 @@ describe('<UserAvatar />', () => {
 - **避免碎片化**: 不要为了单个测试用例就创建新的顶级 `describe` 块
 
 ```typescript
-// ❌ 错误的组织方式：创建过多顶级块
+//  错误的组织方式：创建过多顶级块
 describe('<UserProfile />', () => {
   it('should render user name', () => {});
 });
@@ -158,7 +169,7 @@ describe('UserProfile edge cases', () => {
   it('should handle missing avatar', () => {});
 });
 
-// ✅ 正确的组织方式：合并相关测试
+//  正确的组织方式：合并相关测试
 describe('<UserProfile />', () => {
   it('should render user name', () => {});
 
@@ -214,9 +225,9 @@ describe('<UserProfile />', () => {
 **修复方法**: 更新了测试文件中的 mock 数据结构，使其与最新的 API 响应格式保持一致。具体修改了 `user.test.ts` 中的 `mockUserData` 对象结构。
 ```
 
-## 🎯 测试编写最佳实践
+## 测试编写最佳实践
 
-### Mock 数据策略：追求"低成本的真实性" 📋
+### Mock 数据策略：追求"低成本的真实性"
 
 **核心原则**: 测试数据应默认追求真实性，只有在引入"高昂的测试成本"时才进行简化。
 
@@ -228,10 +239,10 @@ describe('<UserProfile />', () => {
 - **网络请求**：HTTP 调用、数据库连接
 - **系统调用**：获取系统时间、环境变量等
 
-#### ✅ 推荐做法：Mock 依赖，保留真实数据
+#### 推荐做法：Mock 依赖，保留真实数据
 
 ```typescript
-// ✅ 好的做法：Mock I/O 操作，但使用真实的文件内容格式
+//  好的做法：Mock I/O 操作，但使用真实的文件内容格式
 describe('parseContentType', () => {
   beforeEach(() => {
     // Mock 文件读取操作（避免真实 I/O）
@@ -249,7 +260,7 @@ describe('parseContentType', () => {
   });
 });
 
-// ❌ 过度简化：使用不真实的数据
+//  过度简化：使用不真实的数据
 describe('parseContentType', () => {
   it('should detect PDF content type correctly', () => {
     // 这种简化数据没有测试价值
@@ -259,78 +270,116 @@ describe('parseContentType', () => {
 });
 ```
 
-#### 🎯 真实标识符的价值
+#### 真实标识符的价值
 
 ```typescript
-// ✅ 使用真实的提供商标识符
-it('should parse OpenAI model list correctly', () => {
-  const result = parseModelString('openai', '+gpt-4,+gpt-3.5-turbo');
-  expect(result.add).toHaveLength(2);
-  expect(result.add[0].id).toBe('gpt-4');
-});
+// ✅ 使用真实标识符
+const result = parseModelString('openai', '+gpt-4,+gpt-3.5-turbo');
 
-// ❌ 使用占位符标识符（价值较低）
-it('should parse model list correctly', () => {
-  const result = parseModelString('test-provider', '+model1,+model2');
-  expect(result.add).toHaveLength(2);
-  // 这种测试对理解真实场景帮助不大
+// ❌ 使用占位符（价值较低）
+const result = parseModelString('test-provider', '+model1,+model2');
+```
+
+### 现代化Mock技巧：环境设置与Mock方法
+
+**环境设置 + Mock方法结合使用**
+
+客户端代码测试时，推荐使用环境注释配合现代化Mock方法：
+
+```typescript
+/**
+ * @vitest-environment happy-dom  // 提供浏览器API
+ */
+import { beforeEach, vi } from 'vitest';
+
+beforeEach(() => {
+  // 现代方法1：使用vi.stubGlobal替代global.xxx = ...
+  const mockImage = vi.fn().mockImplementation(() => ({
+    addEventListener: vi.fn(),
+    naturalHeight: 600,
+    naturalWidth: 800,
+  }));
+  vi.stubGlobal('Image', mockImage);
+
+  // 现代方法2：使用vi.spyOn保留原功能，只mock特定方法
+  vi.spyOn(URL, 'createObjectURL').mockReturnValue('blob:mock-url');
+  vi.spyOn(URL, 'revokeObjectURL').mockImplementation(() => {});
 });
 ```
 
-### 错误处理测试：测试"行为"而非"文本" ⚠️
+**环境选择优先级**
+
+1. **@vitest-environment happy-dom** (推荐) - 轻量、快速，项目已安装
+2. **@vitest-environment jsdom** - 功能完整，但需要额外安装jsdom包
+3. **不设置环境** - Node.js环境，需要手动mock所有浏览器API
+
+**Mock方法对比**
+
+```typescript
+// ❌ 旧方法：直接操作global对象（类型问题）
+global.Image = mockImage;
+global.URL = { ...global.URL, createObjectURL: mockFn };
+
+// ✅ 现代方法：类型安全的vi API
+vi.stubGlobal('Image', mockImage); // 完全替换全局对象
+vi.spyOn(URL, 'createObjectURL'); // 部分mock，保留其他功能
+```
+
+### 测试覆盖率原则：代码分支优于用例数量
+
+**核心原则**: 优先覆盖所有代码分支，而非编写大量重复用例
+
+```typescript
+// ❌ 过度测试：29个测试用例都验证相同分支
+describe('getImageDimensions', () => {
+  it('should reject .txt files');
+  it('should reject .pdf files');
+  // ... 25个类似测试，都走相同的验证分支
+});
+
+// ✅ 精简测试：4个核心用例覆盖所有分支
+describe('getImageDimensions', () => {
+  it('should return dimensions for valid File object'); // 成功路径 - File
+  it('should return dimensions for valid data URI'); // 成功路径 - String
+  it('should return undefined for invalid inputs'); // 输入验证分支
+  it('should return undefined when image fails to load'); // 错误处理分支
+});
+```
+
+**分支覆盖策略**
+
+1. **成功路径** - 每种输入类型1个测试即可
+2. **边界条件** - 合并类似场景到单个测试
+3. **错误处理** - 测试代表性错误即可
+4. **业务逻辑** - 覆盖所有if/else分支
+
+**合理测试数量**
+
+- 简单工具函数：2-5个测试
+- 复杂业务逻辑：5-10个测试
+- 核心安全功能：适当增加，但避免重复路径
+
+### 错误处理测试：测试"行为"而非"文本"
 
 **核心原则**: 测试应该验证程序在错误发生时的行为是可预测的，而不是验证易变的错误信息文本。
 
-#### ✅ 推荐的错误测试方式
+#### 推荐的错误测试方式
 
 ```typescript
-// ✅ 测试是否抛出错误
-it('should throw error when invalid input provided', () => {
-  expect(() => processInput(null)).toThrow();
-});
+// ✅ 测试错误类型和属性
+expect(() => validateUser({})).toThrow(ValidationError);
+expect(() => processPayment({})).toThrow(
+  expect.objectContaining({
+    code: 'INVALID_PAYMENT_DATA',
+    statusCode: 400,
+  }),
+);
 
-// ✅ 测试错误类型（最推荐）
-it('should throw ValidationError for invalid data', () => {
-  expect(() => validateUser({})).toThrow(ValidationError);
-});
-
-// ✅ 测试错误属性而非消息文本
-it('should throw error with correct error code', () => {
-  expect(() => processPayment({})).toThrow(
-    expect.objectContaining({
-      code: 'INVALID_PAYMENT_DATA',
-      statusCode: 400,
-    }),
-  );
-});
+// ❌ 避免测试具体错误文本
+expect(() => processUser({})).toThrow('用户数据不能为空，请检查输入参数');
 ```
 
-#### ❌ 应避免的做法
-
-```typescript
-// ❌ 过度依赖具体错误信息文本
-it('should throw specific error message', () => {
-  expect(() => processUser({})).toThrow('用户数据不能为空，请检查输入参数');
-  // 这种测试很脆弱，错误文案稍有修改就会失败
-});
-```
-
-#### 🎯 例外情况：何时可以测试错误信息
-
-```typescript
-// ✅ 测试标准 API 错误（这是契约的一部分）
-it('should return proper HTTP error for API', () => {
-  expect(response.statusCode).toBe(400);
-  expect(response.error).toBe('Bad Request');
-});
-
-// ✅ 测试错误信息的关键部分（使用正则）
-it('should include field name in validation error', () => {
-  expect(() => validateField('email', '')).toThrow(/email/i);
-});
-```
-
-### 疑难解答：警惕模块污染 🚨
+### 疑难解答：警惕模块污染
 
 **识别信号**: 当你的测试出现以下"灵异"现象时，优先怀疑模块污染：
 
@@ -341,55 +390,25 @@ it('should include field name in validation error', () => {
 #### 典型场景：动态 Mock 同一模块
 
 ```typescript
-// ❌ 容易出现模块污染的写法
-describe('ConfigService', () => {
-  it('should work in development mode', async () => {
-    vi.doMock('./config', () => ({ isDev: true }));
-    const { getSettings } = await import('./configService'); // 第一次加载
-    expect(getSettings().debugMode).toBe(true);
-  });
-
-  it('should work in production mode', async () => {
-    vi.doMock('./config', () => ({ isDev: false }));
-    const { getSettings } = await import('./configService'); // 可能使用缓存的旧版本！
-    expect(getSettings().debugMode).toBe(false); // ❌ 可能失败
-  });
+// ❌ 问题：动态Mock同一模块
+it('dev mode', async () => {
+  vi.doMock('./config', () => ({ isDev: true }));
+  const { getSettings } = await import('./service'); // 可能使用缓存
 });
 
-// ✅ 使用 resetModules 解决模块污染
-describe('ConfigService', () => {
-  beforeEach(() => {
-    vi.resetModules(); // 清除模块缓存，确保每个测试都是干净的环境
-  });
-
-  it('should work in development mode', async () => {
-    vi.doMock('./config', () => ({ isDev: true }));
-    const { getSettings } = await import('./configService');
-    expect(getSettings().debugMode).toBe(true);
-  });
-
-  it('should work in production mode', async () => {
-    vi.doMock('./config', () => ({ isDev: false }));
-    const { getSettings } = await import('./configService');
-    expect(getSettings().debugMode).toBe(false); // ✅ 测试通过
-  });
+// ✅ 解决：清除模块缓存
+beforeEach(() => {
+  vi.resetModules(); // 确保每个测试都是干净环境
 });
 ```
 
-#### 🔧 排查和解决步骤
+**记住**: `vi.resetModules()` 是解决测试"灵异"失败的终极武器。
 
-1. **识别问题**: 测试失败时，首先问自己："是否有多个测试在 Mock 同一个模块？"
-2. **添加隔离**: 在 `beforeEach` 中添加 `vi.resetModules()`
-3. **验证修复**: 重新运行测试，确认问题解决
-
-**记住**: `vi.resetModules()` 是解决测试"灵异"失败的终极武器，当常规调试方法都无效时，它往往能一针见血地解决问题。
-
-## 📂 测试文件组织
+## 测试文件组织
 
 ### 文件命名约定
 
-- **客户端测试**: `*.test.ts`, `*.test.tsx` (任意位置)
-- **服务端测试**: `src/database/models/**/*.test.ts`, `src/database/server/**/*.test.ts` (限定路径)
+`*.test.ts`, `*.test.tsx` (任意位置)
 
 ### 测试文件组织风格
 
@@ -406,7 +425,10 @@ src/components/Button/
 └── index.test.tsx      # 测试文件
 ```
 
-## 🛠️ 测试调试技巧
+- 也有少数情况会统一放到 `__tests__` 文件夹， 例如 `packages/database/src/models/__tests__`
+- 测试使用的辅助文件放到 fixtures 文件夹
+
+## 测试调试技巧
 
 ### 测试调试步骤
 
@@ -415,40 +437,28 @@ src/components/Button/
 3. **分析错误**: 仔细阅读错误信息、堆栈跟踪和最近的文件修改记录
 4. **添加调试**: 在测试中添加 `console.log` 了解执行流程
 
-### TypeScript 类型处理 📝
+### TypeScript 类型处理
 
 在测试中，为了提高编写效率和可读性，可以适当放宽 TypeScript 类型检测：
 
-#### ✅ 推荐的类型放宽策略
+#### 推荐的类型放宽策略
 
 ```typescript
-// ✅ 使用非空断言访问测试中确定存在的属性
+//  使用非空断言访问测试中确定存在的属性
 const result = await someFunction();
 expect(result!.data).toBeDefined();
 expect(result!.status).toBe('success');
 
-// ✅ 使用 any 类型简化复杂的 Mock 设置
+//  使用 any 类型简化复杂的 Mock 设置
 const mockStream = new ReadableStream() as any;
 mockStream.toReadableStream = () => mockStream;
 
-// ✅ 使用中括号访问私有属性和方法（推荐）
-class MyClass {
-  private _cache = new Map();
-  private getFromCache(key: string) { /* ... */ }
-}
-
-const instance = new MyClass();
-
-// 推荐：使用中括号访问私有成员
-await instance['getFromCache']('test-key');
-expect(instance['_cache'].size).toBe(1);
-
-// 避免：使用 as any 访问私有成员
-await (instance as any).getFromCache('test-key');  // ❌ 不推荐
-expect((instance as any)._cache.size).toBe(1);     // ❌ 不推荐
+//  访问私有成员
+await instance['getFromCache']('key'); // 推荐中括号
+await (instance as any).getFromCache('key'); // 避免as any
 ```
 
-#### 🎯 适用场景
+#### 适用场景
 
 - **Mock 对象**: 对于测试用的 Mock 数据，使用 `as any` 避免复杂的类型定义
 - **第三方库**: 处理复杂的第三方库类型时，适当使用 `any` 提高效率
@@ -456,44 +466,30 @@ expect((instance as any)._cache.size).toBe(1);     // ❌ 不推荐
 - **私有成员访问**: 优先使用中括号 `instance['privateMethod']()` 而不是 `(instance as any).privateMethod()`
 - **临时调试**: 快速编写测试时，先用 `any` 保证功能，后续可选择性地优化类型
 
-#### ⚠️ 注意事项
+#### 注意事项
 
 - **适度使用**: 不要过度依赖 `any`，核心业务逻辑的类型仍应保持严格
 - **私有成员访问优先级**: 中括号访问 > `as any` 转换，保持更好的类型安全性
 - **文档说明**: 对于使用 `any` 的复杂场景，添加注释说明原因
 - **测试覆盖**: 确保即使使用了 `any`，测试仍能有效验证功能正确性
 
-### 检查最近修改记录 🔍
+### 检查最近修改记录
 
-系统性地检查相关文件的修改历史是问题定位的关键步骤。
+**核心原则**：测试突然失败时，优先检查最近的代码修改。
 
-#### 三步检查法
-
-**Step 1: 查看当前状态**
+#### 快速检查方法
 
 ```bash
-git status                               # 查看未提交的修改
-git diff path/to/component.test.ts | cat # 查看测试文件修改
-git diff path/to/component.ts | cat      # 查看实现文件修改
+git status                  # 查看当前修改状态
+git diff HEAD -- '*.test.*' # 检查测试文件改动
+git diff main...HEAD        # 对比主分支差异
+gh pr diff                  # 查看PR中的所有改动
 ```
 
-**Step 2: 查看提交历史**
+#### 常见原因与解决
 
-```bash
-git log --pretty=format:"%h %ad %s" --date=relative -3 path/to/component.ts | cat
-```
-
-**Step 3: 查看具体修改内容**
-
-```bash
-git show HEAD -- path/to/component.ts | cat # 查看最新提交的修改
-```
-
-#### 时间相关性判断
-
-- **24小时内的提交**: 🔴 **高度相关** - 很可能是直接原因
-- **1-7天内的提交**: 🟡 **中等相关** - 需要仔细分析
-- **超过1周的提交**: ⚪ **低相关性** - 除非重大重构
+- **最新提交引入bug** → 检查并修复实现代码
+- **分支代码滞后** → `git rebase main` 同步主分支
 
 ## 特殊场景的测试
 
@@ -502,9 +498,9 @@ git show HEAD -- path/to/component.ts | cat # 查看最新提交的修改
 - [Electron IPC 接口测试策略](mdc:./electron-ipc-test.mdc)
 - [数据库 Model 测试指南](mdc:./db-model-test.mdc)
 
-## 🎯 核心要点
+## 核心要点
 
-- **命令格式**: 使用 `npx vitest run --config [config-file]` 并指定文件过滤
+- **命令格式**: 使用 `bunx vitest run --silent='passed-only'` 并指定文件过滤
 - **修复原则**: 失败1-2次后寻求帮助，测试命名关注行为而非实现细节
 - **调试流程**: 复现 → 分析 → 假设 → 修复 → 验证 → 总结
 - **文件组织**: 优先在现有 `describe` 块中添加测试，避免创建冗余顶级块

.env.example

@@ -4,6 +4,14 @@
 # Specify your API Key selection method, currently supporting `random` and `turn`.
 # API_KEY_SELECT_MODE=random
 
+########################################
+########### Security Settings ###########
+########################################
+
+# Control Content Security Policy headers
+# Set to '1' to enable X-Frame-Options and Content-Security-Policy headers
+# Default is '0' (enabled)
+# ENABLED_CSP=1
 
 ########################################
 ########## AI Provider Service #########
@@ -153,6 +161,17 @@ OPENAI_API_KEY=sk-xxxxxxxxx
 
 # AIHUBMIX_API_KEY=sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
 
+### BFL ###
+
+# BFL_API_KEY=bfl-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
+
+### FAL ###
+
+# FAL_API_KEY=fal-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
+
+### Nebius ###
+
+# NEBIUS_API_KEY=sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
 
 ########################################
 ############ Market Service ############

.env.example.development

@@ -0,0 +1,122 @@
+# LobeChat Development Server Configuration
+# This file contains environment variables for both LobeChat server mode and Docker compose setup
+
+COMPOSE_FILE="docker-compose.development.yml"
+
+# ⚠️⚠️⚠️ DO NOT USE THE SECRETS BELOW IN PRODUCTION!
+UNSAFE_SECRET="ww+0igxjGRAAR/eTNFQ55VmhQB5KE5trFZseuntThJs="
+UNSAFE_PASSWORD="CHANGE_THIS_PASSWORD_IN_PRODUCTION"
+
+# Core Server Configuration
+# Service mode - set to 'server' for server-side deployment
+NEXT_PUBLIC_SERVICE_MODE=server
+
+# Service Ports Configuration
+LOBE_PORT=3010
+
+# Application URL - the base URL where LobeChat will be accessible
+APP_URL=http://localhost:${LOBE_PORT}
+
+# Secret key for encrypting vault data (generate with: openssl rand -base64 32)
+KEY_VAULTS_SECRET=${UNSAFE_SECRET}
+
+# Database Configuration
+# Database name for LobeChat
+LOBE_DB_NAME=lobechat
+
+# PostgreSQL password
+POSTGRES_PASSWORD=${UNSAFE_PASSWORD}
+
+# PostgreSQL database connection URL
+DATABASE_URL=postgresql://postgres:${POSTGRES_PASSWORD}@localhost:5432/${LOBE_DB_NAME}
+
+# Database driver type
+DATABASE_DRIVER=node
+
+# Authentication Configuration
+# Enable NextAuth authentication
+NEXT_PUBLIC_ENABLE_NEXT_AUTH=1
+
+# NextAuth secret for JWT signing (generate with: openssl rand -base64 32)
+NEXT_AUTH_SECRET=${UNSAFE_SECRET}
+
+NEXTAUTH_URL=${APP_URL}
+
+# Authentication URL
+AUTH_URL=${APP_URL}/api/auth
+
+# SSO providers configuration - using Casdoor for development
+NEXT_AUTH_SSO_PROVIDERS=casdoor
+
+# Casdoor Configuration
+# Casdoor service port
+CASDOOR_PORT=8000
+
+# Casdoor OIDC issuer URL
+AUTH_CASDOOR_ISSUER=http://localhost:${CASDOOR_PORT}
+
+# Casdoor application client ID
+AUTH_CASDOOR_ID=a387a4892ee19b1a2249 # DO NOT USE IN PROD
+
+# Casdoor application client secret
+AUTH_CASDOOR_SECRET=dbf205949d704de81b0b5b3603174e23fbecc354 # DO NOT USE IN PROD
+
+# Origin URL for Casdoor internal configuration
+origin=http://localhost:${CASDOOR_PORT}
+
+# MinIO Storage Configuration
+# MinIO service port
+MINIO_PORT=9000
+
+# MinIO root user (admin username)
+MINIO_ROOT_USER=admin
+
+# MinIO root password
+MINIO_ROOT_PASSWORD=${UNSAFE_PASSWORD}
+
+# MinIO bucket for LobeChat files
+MINIO_LOBE_BUCKET=lobe
+
+# S3/MinIO Configuration for LobeChat
+# S3/MinIO access key ID
+S3_ACCESS_KEY_ID=${MINIO_ROOT_USER}
+
+# S3/MinIO secret access key
+S3_SECRET_ACCESS_KEY=${MINIO_ROOT_PASSWORD}
+
+# S3/MinIO endpoint URL
+S3_ENDPOINT=http://localhost:${MINIO_PORT}
+
+# S3 bucket name for storing files
+S3_BUCKET=${MINIO_LOBE_BUCKET}
+
+# Public domain for S3 file access
+S3_PUBLIC_DOMAIN=http://localhost:${MINIO_PORT}
+
+# Enable path-style S3 requests (required for MinIO)
+S3_ENABLE_PATH_STYLE=1
+
+# Disable S3 ACL setting (for MinIO compatibility)
+S3_SET_ACL=0
+
+# Use base64 encoding for LLM vision images
+LLM_VISION_IMAGE_USE_BASE64=1
+
+# Search Service Configuration
+# SearXNG search engine URL
+SEARXNG_URL=http://searxng:8080
+
+# Development Options
+# Uncomment to skip authentication during development
+
+# Proxy Configuration (Optional)
+# Uncomment if you need proxy support (e.g., for GitHub auth or API access)
+# HTTP_PROXY=http://localhost:7890
+# HTTPS_PROXY=http://localhost:7890
+
+# AI Model Configuration (Optional)
+# Add your AI model API keys and configurations here
+# ⚠️  WARNING: Never commit real API keys to version control!
+# OPENAI_API_KEY=sk-NEVER_USE_REAL_API_KEYS_IN_CONFIG_FILES
+# OPENAI_PROXY_URL=https://api.openai.com/v1
+# OPENAI_MODEL_LIST=...

.eslintignore

@@ -29,3 +29,20 @@ logs
 # misc
 # add other ignore file below
 .next
+
+# temporary directories
+tmp
+temp
+.temp
+.local
+docs/.local
+
+# cache directories
+.cache
+
+# AI coding tools directories
+.claude
+.serena
+
+# MCP tools
+/.serena/**

.github/PULL_REQUEST_TEMPLATE.md

@@ -8,6 +8,7 @@
 - [ ] 💄 style
 - [ ] 👷 build
 - [ ] ⚡️ perf
+- [ ] ✅ test
 - [ ] 📝 docs
 - [ ] 🔨 chore
 

.github/scripts/docker-pr-comment.js

@@ -0,0 +1,89 @@
+/**
+ * Generate or update PR comment with Docker build info
+ */
+module.exports = async ({
+  github,
+  context,
+  dockerMetaJson,
+  image,
+  version,
+  dockerhubUrl,
+  platforms,
+}) => {
+  const COMMENT_IDENTIFIER = '<!-- DOCKER-BUILD-COMMENT -->';
+
+  const parseTags = () => {
+    try {
+      if (dockerMetaJson) {
+        const parsed = JSON.parse(dockerMetaJson);
+        if (Array.isArray(parsed.tags) && parsed.tags.length > 0) {
+          return parsed.tags;
+        }
+      }
+    } catch (e) {
+      // ignore parsing error, fallback below
+    }
+    if (image && version) {
+      return [`${image}:${version}`];
+    }
+    return [];
+  };
+
+  const generateCommentBody = () => {
+    const tags = parseTags();
+    const buildTime = new Date().toISOString();
+
+    // Use the first tag as the main version
+    const mainTag = tags.length > 0 ? tags[0] : `${image}:${version}`;
+    const tagVersion = mainTag.includes(':') ? mainTag.split(':')[1] : version;
+
+    return [
+      COMMENT_IDENTIFIER,
+      '',
+      '### 🐳 Database Docker Build Completed!',
+      `**Version**: \`${tagVersion || 'N/A'}\``,
+      `**Build Time**: \`${buildTime}\``,
+      '',
+      dockerhubUrl ? `🔗 View all tags on Docker Hub: ${dockerhubUrl}` : '',
+      '',
+      '### Pull Image',
+      'Download the Docker image to your local machine:',
+      '',
+      '```bash',
+      `docker pull ${mainTag}`,
+      '```',
+      '> [!IMPORTANT]',
+      '> This build is for testing and validation purposes.',
+    ]
+      .filter(Boolean)
+      .join('\n');
+  };
+
+  const body = generateCommentBody();
+
+  // List comments on the PR
+  const { data: comments } = await github.rest.issues.listComments({
+    issue_number: context.issue.number,
+    owner: context.repo.owner,
+    repo: context.repo.repo,
+  });
+
+  const existing = comments.find((c) => c.body && c.body.includes(COMMENT_IDENTIFIER));
+  if (existing) {
+    await github.rest.issues.updateComment({
+      comment_id: existing.id,
+      owner: context.repo.owner,
+      repo: context.repo.repo,
+      body,
+    });
+    return { updated: true, id: existing.id };
+  }
+
+  const result = await github.rest.issues.createComment({
+    issue_number: context.issue.number,
+    owner: context.repo.owner,
+    repo: context.repo.repo,
+    body,
+  });
+  return { updated: false, id: result.data.id };
+};

.github/workflows/auto-i18n.yml

@@ -21,7 +21,7 @@ jobs:
           git config --global user.name "lobehubbot"
           git config --global user.email "i@lobehub.com"
 
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
         with:
           ref: ${{ github.event.pull_request.head.ref }}
 

.github/workflows/claude-code-review.yml

@@ -1,78 +0,0 @@
-name: Claude Code Review
-
-on:
-  pull_request:
-    types: [opened, synchronize]
-    # Optional: Only run on specific file changes
-    # paths:
-    #   - "src/**/*.ts"
-    #   - "src/**/*.tsx"
-    #   - "src/**/*.js"
-    #   - "src/**/*.jsx"
-
-jobs:
-  claude-review:
-    # Optional: Filter by PR author
-    # if: |
-    #   github.event.pull_request.user.login == 'external-contributor' ||
-    #   github.event.pull_request.user.login == 'new-developer' ||
-    #   github.event.pull_request.author_association == 'FIRST_TIME_CONTRIBUTOR'
-
-    runs-on: ubuntu-latest
-    permissions:
-      contents: read
-      pull-requests: read
-      issues: read
-      id-token: write
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 1
-
-      - name: Run Claude Code Review
-        id: claude-review
-        uses: anthropics/claude-code-action@beta
-        with:
-          claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
-
-          # Optional: Specify model (defaults to Claude Sonnet 4, uncomment for Claude Opus 4)
-          model: 'claude-opus-4-1-20250805'
-          allowed_bots: 'bot'
-
-          # Direct prompt for automated review (no @claude mention needed)
-          direct_prompt: |
-            Please review this pull request and provide feedback on:
-            - Code quality and best practices
-            - Potential bugs or issues
-            - Performance considerations
-            - Security concerns
-            - Test coverage
-
-            Be constructive and helpful in your feedback.
-
-          # Optional: Use sticky comments to make Claude reuse the same comment on subsequent pushes to the same PR
-          # use_sticky_comment: true
-
-          # Optional: Customize review based on file types
-          # direct_prompt: |
-          #   Review this PR focusing on:
-          #   - For TypeScript files: Type safety and proper interface usage
-          #   - For API endpoints: Security, input validation, and error handling
-          #   - For React components: Performance, accessibility, and best practices
-          #   - For tests: Coverage, edge cases, and test quality
-
-          # Optional: Different prompts for different authors
-          # direct_prompt: |
-          #   ${{ github.event.pull_request.author_association == 'FIRST_TIME_CONTRIBUTOR' &&
-          #   'Welcome! Please review this PR from a first-time contributor. Be encouraging and provide detailed explanations for any suggestions.' ||
-          #   'Please provide a thorough code review focusing on our coding standards and best practices.' }}
-
-          # Optional: Add specific tools for running tests or linting
-          allowed_tools: 'Bash(bun run:*),Bash(pnpm run:*),Bash(npm run:*),Bash(npx vitest:*),Bash(rg:*),Bash(find:*),Bash(sed:*),Bash(grep:*),Bash(awk:*),Bash(wc:*),Bash(xargs:*)'
-
-          # Optional: Skip review for certain conditions
-          # if: |
-          #   !contains(github.event.pull_request.title, '[skip-review]') &&
-          #   !contains(github.event.pull_request.title, '[WIP]')

.github/workflows/claude.yml

@@ -26,7 +26,7 @@ jobs:
       actions: read # Required for Claude to read CI results on PRs
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
         with:
           fetch-depth: 1
 
@@ -41,7 +41,7 @@ jobs:
             actions: read
 
           # Optional: Specify model (defaults to Claude Sonnet 4, uncomment for Claude Opus 4)
-          model: 'claude-opus-4-1-20250805'
+          # model: 'claude-opus-4-1-20250805'
           allowed_bots: 'bot'
 
           # Optional: Customize the trigger phrase (default: @claude)
@@ -51,7 +51,7 @@ jobs:
           # assignee_trigger: "claude-bot"
 
           # Optional: Allow Claude to run specific commands
-          allowed_tools: 'Bash(bun run:*),Bash(pnpm run:*),Bash(npm run:*),Bash(npx vitest:*),Bash(rg:*),Bash(find:*),Bash(sed:*),Bash(grep:*),Bash(awk:*),Bash(wc:*),Bash(xargs:*)'
+          allowed_tools: 'Bash(bun run:*),Bash(pnpm run:*),Bash(npm run:*),Bash(npx:*),Bash(bunx:*),Bash(vitest:*),Bash(rg:*),Bash(find:*),Bash(sed:*),Bash(grep:*),Bash(awk:*),Bash(wc:*),Bash(xargs:*)'
 
           # Optional: Add custom instructions for Claude to customize its behavior for your project
           # custom_instructions: |

.github/workflows/desktop-pr-build.yml

@@ -23,7 +23,7 @@ jobs:
     runs-on: ubuntu-latest # 只在 ubuntu 上运行一次检查
     steps:
       - name: Checkout base
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
         with:
           fetch-depth: 0
 
@@ -56,7 +56,7 @@ jobs:
       # 输出版本信息，供后续 job 使用
       version: ${{ steps.set_version.outputs.version }}
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
         with:
           fetch-depth: 0
 
@@ -95,7 +95,7 @@ jobs:
       matrix:
         os: [macos-latest, windows-2025, ubuntu-latest]
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
         with:
           fetch-depth: 0
 
@@ -183,6 +183,10 @@ jobs:
             apps/desktop/release/*.zip*
             apps/desktop/release/*.exe*
             apps/desktop/release/*.AppImage
+            apps/desktop/release/*.deb*
+            apps/desktop/release/*.snap*
+            apps/desktop/release/*.rpm*
+            apps/desktop/release/*.tar.gz*
           retention-days: 5
 
   publish-pr:
@@ -196,7 +200,7 @@ jobs:
     outputs:
       artifact_path: ${{ steps.set_path.outputs.path }}
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
         with:
           fetch-depth: 0
 
@@ -245,6 +249,10 @@ jobs:
             release/*.zip*
             release/*.exe*
             release/*.AppImage
+            release/*.deb*
+            release/*.snap*
+            release/*.rpm*
+            release/*.tar.gz*
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
 

.github/workflows/docker-database.yml

@@ -39,7 +39,7 @@ jobs:
           echo "PLATFORM_PAIR=${platform//\//-}" >> $GITHUB_ENV
 
       - name: Checkout base
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
         with:
           fetch-depth: 0
 
@@ -80,7 +80,7 @@ jobs:
 
       - name: Build and export
         id: build
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v6
         with:
           platforms: ${{ matrix.platform }}
           context: .
@@ -111,7 +111,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout base
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
         with:
           fetch-depth: 0
 
@@ -159,3 +159,24 @@ jobs:
       - name: Inspect image
         run: |
           docker buildx imagetools inspect ${{ env.REGISTRY_IMAGE }}:${{ steps.meta.outputs.version }}
+
+
+      - name: Comment on PR with Docker build info
+        if: github.event_name == 'pull_request'
+        uses: actions/github-script@v7
+        with:
+          github-token: ${{ secrets.GITHUB_TOKEN }}
+          script: |
+            const prComment = require('${{ github.workspace }}/.github/scripts/docker-pr-comment.js');
+            const result = await prComment({
+              github,
+              context,
+              dockerMetaJson: ${{ toJSON(steps.meta.outputs.json) }},
+              image: "${{ env.REGISTRY_IMAGE }}",
+              version: "${{ steps.meta.outputs.version }}",
+              dockerhubUrl: "https://hub.docker.com/r/${{ env.REGISTRY_IMAGE }}/tags",
+              platforms: "linux/amd64, linux/arm64",
+            });
+            core.info(`Status: ${result.updated ? 'Updated' : 'Created'}, ID: ${result.id}`);
+
+

.github/workflows/docker-pglite.yml

@@ -39,7 +39,7 @@ jobs:
           echo "PLATFORM_PAIR=${platform//\//-}" >> $GITHUB_ENV
 
       - name: Checkout base
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
         with:
           fetch-depth: 0
 
@@ -80,7 +80,7 @@ jobs:
 
       - name: Build and export
         id: build
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v6
         with:
           platforms: ${{ matrix.platform }}
           context: .
@@ -111,7 +111,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout base
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
         with:
           fetch-depth: 0
 

.github/workflows/docker.yml

@@ -39,7 +39,7 @@ jobs:
           echo "PLATFORM_PAIR=${platform//\//-}" >> $GITHUB_ENV
 
       - name: Checkout base
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
         with:
           fetch-depth: 0
 
@@ -80,7 +80,7 @@ jobs:
 
       - name: Build and export
         id: build
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v6
         with:
           platforms: ${{ matrix.platform }}
           context: .
@@ -111,7 +111,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout base
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
         with:
           fetch-depth: 0
 

.github/workflows/lighthouse.yml

@@ -42,12 +42,12 @@ jobs:
           echo "BRANCH=$BRANCH" >> $GITHUB_ENV
         env:
           REPO_BRANCH: ${{ matrix.REPO_BRANCH || env.REPO_BRANCH }}
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
         with:
           repository: ${{ env.REPOSITORY }}
           token: ${{ secrets[matrix.TOKEN_NAME] || secrets[env.TOKEN_NAME] }}
           ref: ${{ env.BRANCH }}
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
         with:
           repository: 'myactionway/lighthouse-badges'
           path: temp_lighthouse_badges_nested

.github/workflows/release-desktop-beta.yml

@@ -19,7 +19,7 @@ jobs:
     runs-on: ubuntu-latest # 只在 ubuntu 上运行一次检查
     steps:
       - name: Checkout base
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
         with:
           fetch-depth: 0
 
@@ -47,7 +47,7 @@ jobs:
       version: ${{ steps.set_version.outputs.version }}
       is_pr_build: ${{ steps.set_version.outputs.is_pr_build }}
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
         with:
           fetch-depth: 0
 
@@ -82,7 +82,7 @@ jobs:
       matrix:
         os: [macos-latest, windows-2025, ubuntu-latest]
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
         with:
           fetch-depth: 0
 
@@ -165,6 +165,10 @@ jobs:
             apps/desktop/release/*.zip*
             apps/desktop/release/*.exe*
             apps/desktop/release/*.AppImage
+            apps/desktop/release/*.deb*
+            apps/desktop/release/*.snap*
+            apps/desktop/release/*.rpm*
+            apps/desktop/release/*.tar.gz*
           retention-days: 5
 
   # 正式版发布 job
@@ -201,5 +205,9 @@ jobs:
             release/*.zip*
             release/*.exe*
             release/*.AppImage
+            release/*.deb*
+            release/*.snap*
+            release/*.rpm*
+            release/*.tar.gz*
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/release.yml

@@ -11,16 +11,20 @@ jobs:
 
     services:
       postgres:
-        image: pgvector/pgvector:pg16
+        image: paradedb/paradedb:latest
         env:
           POSTGRES_PASSWORD: postgres
         options: >-
           --health-cmd pg_isready --health-interval 10s --health-timeout 5s  --health-retries 5
+
+
         ports:
           - 5432:5432
 
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
+        with:
+          token: ${{ secrets.GH_TOKEN }}
 
       - name: Setup Node.js
         uses: actions/setup-node@v4
@@ -38,8 +42,8 @@ jobs:
       - name: Lint
         run: bun run lint
 
-      - name: Test Server Coverage
-        run: bun run test-server:coverage
+      - name: Test Database Coverage
+        run: bun run --filter @lobechat/database test
         env:
           DATABASE_TEST_URL: postgresql://postgres:postgres@localhost:5432/postgres
           DATABASE_DRIVER: node
@@ -48,8 +52,8 @@ jobs:
           S3_PUBLIC_DOMAIN: https://example.com
           APP_URL: https://home.com
 
-      - name: Test App Coverage
-        run: bun run test-app:coverage
+      - name: Test App
+        run: bun run test-app
 
       - name: Release
         run: bun run release

.github/workflows/sync-database-schema.yml

@@ -11,7 +11,7 @@ jobs:
   build:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
 
       - name: Install dbdocs
         run: sudo npm install -g dbdocs

.github/workflows/sync.yml

@@ -17,7 +17,7 @@ jobs:
     if: ${{ github.event.repository.fork }}
 
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
 
       - name: Clean issue notice
         uses: actions-cool/issues-helper@v3

.github/workflows/test.yml

@@ -2,8 +2,115 @@ name: Test CI
 
 on: [push, pull_request]
 
+permissions:
+  contents: read
+
 jobs:
-  test:
+  # Package tests - using each package's own test script
+  test-intenral-packages:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        package: [file-loaders, prompts, model-runtime, web-crawler, electron-server-ipc, utils]
+
+    name: Test package ${{ matrix.package }}
+
+    steps:
+      - uses: actions/checkout@v5
+
+      - name: Setup Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version: 22
+
+      - name: Install bun
+        uses: oven-sh/setup-bun@v1
+        with:
+          bun-version: ${{ secrets.BUN_VERSION }}
+
+      - name: Install deps
+        run: bun i
+
+      - name: Test ${{ matrix.package }} package with coverage
+        run: bun run --filter @lobechat/${{ matrix.package }} test:coverage
+
+      - name: Upload ${{ matrix.package }} coverage to Codecov
+        uses: codecov/codecov-action@v4
+        with:
+          token: ${{ secrets.CODECOV_TOKEN }}
+          files: ./packages/${{ matrix.package }}/coverage/lcov.info
+          flags: packages/${{ matrix.package }}
+
+  test-packages:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        package: [model-bank]
+
+    name: Test package ${{ matrix.package }}
+
+    steps:
+      - uses: actions/checkout@v5
+
+      - name: Setup Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version: 22
+
+      - name: Install bun
+        uses: oven-sh/setup-bun@v1
+        with:
+          bun-version: ${{ secrets.BUN_VERSION }}
+
+      - name: Install deps
+        run: bun i
+
+      - name: Test ${{ matrix.package }} package with coverage
+        run: bun run --filter ${{ matrix.package }} test:coverage
+
+      - name: Upload ${{ matrix.package }} coverage to Codecov
+        uses: codecov/codecov-action@v4
+        with:
+          token: ${{ secrets.CODECOV_TOKEN }}
+          files: ./packages/${{ matrix.package }}/coverage/lcov.info
+          flags: packages/${{ matrix.package }}
+
+
+  # App tests
+  test-website:
+    name: Test Website
+
+    runs-on: ubuntu-latest
+
+    steps:
+      - uses: actions/checkout@v5
+
+      - name: Setup Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version: 22
+
+      - name: Install bun
+        uses: oven-sh/setup-bun@v1
+        with:
+          bun-version: ${{ secrets.BUN_VERSION }}
+
+      - name: Install deps
+        run: bun i
+
+      - name: Test App Coverage
+        run: bun run test-app:coverage
+
+      - name: Upload App Coverage to Codecov
+        uses: codecov/codecov-action@v4
+        with:
+          token: ${{ secrets.CODECOV_TOKEN }}
+          files: ./coverage/app/lcov.info
+          flags: app
+
+  test-databsae:
+    name: Test Database
+
     runs-on: ubuntu-latest
 
     services:
@@ -19,7 +126,7 @@ jobs:
           - 5432:5432
 
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
 
       - name: Setup Node.js
         uses: actions/setup-node@v4
@@ -37,8 +144,15 @@ jobs:
       - name: Lint
         run: bun run lint
 
-      - name: Test Server Coverage
-        run: bun run test-server:coverage
+      - name: Test Client DB
+        run: bun run --filter @lobechat/database test:client-db
+        env:
+          KEY_VAULTS_SECRET: LA7n9k3JdEcbSgml2sxfw+4TV1AzaaFU5+R176aQz4s=
+          S3_PUBLIC_DOMAIN: https://example.com
+          APP_URL: https://home.com
+
+      - name: Test Coverage
+        run: bun run --filter @lobechat/database test:coverage
         env:
           DATABASE_TEST_URL: postgresql://postgres:postgres@localhost:5432/postgres
           DATABASE_DRIVER: node
@@ -47,19 +161,9 @@ jobs:
           S3_PUBLIC_DOMAIN: https://example.com
           APP_URL: https://home.com
 
-      - name: Upload Server coverage to Codecov
+      - name: Upload Database coverage to Codecov
         uses: codecov/codecov-action@v4
         with:
           token: ${{ secrets.CODECOV_TOKEN }}
-          files: ./coverage/server/lcov.info
-          flags: server
-
-      - name: Test App Coverage
-        run: bun run test-app:coverage
-
-      - name: Upload App Coverage to Codecov
-        uses: codecov/codecov-action@v4
-        with:
-          token: ${{ secrets.CODECOV_TOKEN }}
-          files: ./coverage/app/lcov.info
-          flags: app
+          files: ./packages/database/coverage/lcov.info
+          flags: database

.github/workflows/wiki-sync.yml

@@ -1,19 +0,0 @@
-name: Wiki Sync
-
-on:
-  workflow_dispatch:
-  push:
-    paths:
-      - 'docs/wiki/**'
-    branches:
-      - main
-
-jobs:
-  update-wiki:
-    runs-on: ubuntu-latest
-    name: Wiki sync
-    steps:
-      - uses: OrlovM/Wiki-Action@v1
-        with:
-          path: 'docs/wiki'
-          token: ${{ secrets.GH_TOKEN }}

.gitignore

@@ -104,6 +104,9 @@ vertex-ai-key.json
 
 CLAUDE.local.md
 
+# MCP tools
+.serena/**
+
 # Misc
 ./packages/lobe-ui
 *.ppt*

.i18nrc.js

@@ -25,7 +25,7 @@ module.exports = defineConfig({
   ],
   temperature: 0,
   saveImmediately: true,
-  modelName: 'gpt-5-mini',
+  modelName: 'gpt-4.1-mini',
   experimental: {
     jsonMode: true,
   },

.npmrc

@@ -4,6 +4,9 @@ resolution-mode=highest
 ignore-workspace-root-check=true
 enable-pre-post-scripts=true
 
+# Load dotenv files for all the npm scripts
+node-options="--require dotenv-expand/config"
+
 public-hoist-pattern[]=*@umijs/lint*
 public-hoist-pattern[]=*changelog*
 public-hoist-pattern[]=*commitlint*

.prettierignore

@@ -5,7 +5,6 @@
 .DS_Store
 .editorconfig
 .idea
-.vscode
 .history
 .temp
 .env.local

.stylelintignore

@@ -0,0 +1,39 @@
+# Stylelintignore for LobeHub
+################################################################
+
+# dependencies
+node_modules
+
+# ci
+coverage
+.coverage
+
+# production
+dist
+es
+lib
+logs
+
+# framework specific
+.next
+.umi
+.umi-production
+.umi-test
+.dumi/tmp*
+
+# temporary directories
+tmp
+temp
+.temp
+.local
+docs/.local
+
+# cache directories
+.cache
+
+# AI coding tools directories
+.claude
+.serena
+
+# MCP tools
+/.serena/**

.vscode/settings.json

@@ -1,89 +1,93 @@
 {
-    "npm.packageManager": "pnpm",
-    // don't show errors, but fix when save and git pre commit
-    "eslint.rules.customizations": [
-        { "rule": "import/order", "severity": "off" },
-        { "rule": "prettier/prettier", "severity": "off" },
-        { "rule": "react/jsx-sort-props", "severity": "off" },
-        { "rule": "sort-keys-fix/sort-keys-fix", "severity": "off" },
-        { "rule": "typescript-sort-keys/interface", "severity": "off" }
-    ],
-    "stylelint.validate": [
-        "css",
-        "postcss",
-        // make stylelint work with tsx antd-style css template string
-        "typescriptreact"
-    ],
-    "search.exclude": {
-        "**/node_modules": true,
-        // useless to search this big folder
-        "locales": true
-    },
-    "vitest.maximumConfigs": 6,
-    "workbench.editor.customLabels.patterns": {
-        "**/app/**/[[]*[]]/[[]*[]]/page.tsx": "${dirname(2)}/${dirname(1)}/${dirname} • page component",
-        "**/app/**/[[]*[]]/page.tsx": "${dirname(1)}/${dirname} • page component",
-        "**/app/**/page.tsx": "${dirname} • page component",
+  "editor.codeActionsOnSave": {
+    "source.addMissingImports": "explicit",
+    "source.fixAll.eslint": "explicit",
+    "source.fixAll.stylelint": "explicit"
+  },
+  "editor.formatOnSave": true,
+  // don't show errors, but fix when save and git pre commit
+  "eslint.rules.customizations": [
+    { "rule": "import/order", "severity": "off" },
+    { "rule": "prettier/prettier", "severity": "off" },
+    { "rule": "react/jsx-sort-props", "severity": "off" },
+    { "rule": "sort-keys-fix/sort-keys-fix", "severity": "off" },
+    { "rule": "typescript-sort-keys/interface", "severity": "off" }
+  ],
+  "eslint.validate": [
+    "javascript",
+    "javascriptreact",
+    "typescript",
+    "typescriptreact",
+    // support mdx
+    "mdx"
+  ],
+  "npm.packageManager": "pnpm",
+  "search.exclude": {
+    "**/node_modules": true,
+    // useless to search this big folder
+    "locales": true
+  },
+  "stylelint.validate": [
+    "css",
+    "postcss",
+    // make stylelint work with tsx antd-style css template string
+    "typescriptreact"
+  ],
+  "vitest.maximumConfigs": 20,
+  "workbench.editor.customLabels.patterns": {
+    "**/app/**/[[]*[]]/[[]*[]]/page.tsx": "${dirname(2)}/${dirname(1)}/${dirname} • page component",
+    "**/app/**/[[]*[]]/page.tsx": "${dirname(1)}/${dirname} • page component",
+    "**/app/**/page.tsx": "${dirname} • page component",
 
-        "**/app/**/[[]*[]]/[[]*[]]/layout.tsx": "${dirname(2)}/${dirname(1)}/${dirname} • page layout",
-        "**/app/**/[[]*[]]/layout.tsx": "${dirname(1)}/${dirname} • page layout",
-        "**/app/**/layout.tsx": "${dirname} • page layout",
+    "**/app/**/[[]*[]]/[[]*[]]/layout.tsx": "${dirname(2)}/${dirname(1)}/${dirname} • page layout",
+    "**/app/**/[[]*[]]/layout.tsx": "${dirname(1)}/${dirname} • page layout",
+    "**/app/**/layout.tsx": "${dirname} • page layout",
 
-        "**/app/**/[[]*[]]/[[]*[]]/default.tsx": "${dirname(2)}/${dirname(1)}/${dirname} • slot default",
-        "**/app/**/[[]*[]]/default.tsx": "${dirname(1)}/${dirname} • slot default",
-        "**/app/**/default.tsx": "${dirname} • slot default",
+    "**/app/**/[[]*[]]/[[]*[]]/default.tsx": "${dirname(2)}/${dirname(1)}/${dirname} • slot default",
+    "**/app/**/[[]*[]]/default.tsx": "${dirname(1)}/${dirname} • slot default",
+    "**/app/**/default.tsx": "${dirname} • slot default",
 
-        "**/app/**/[[]*[]]/[[]*[]]/error.tsx": "${dirname(2)}/${dirname(1)}/${dirname} • error component",
-        "**/app/**/[[]*[]]/error.tsx": "${dirname(1)}/${dirname} • error component",
-        "**/app/**/error.tsx": "${dirname} • error component",
+    "**/app/**/[[]*[]]/[[]*[]]/error.tsx": "${dirname(2)}/${dirname(1)}/${dirname} • error component",
+    "**/app/**/[[]*[]]/error.tsx": "${dirname(1)}/${dirname} • error component",
+    "**/app/**/error.tsx": "${dirname} • error component",
 
-        "**/app/**/[[]*[]]/[[]*[]]/loading.tsx": "${dirname(2)}/${dirname(1)}/${dirname} • loading component",
-        "**/app/**/[[]*[]]/loading.tsx": "${dirname(1)}/${dirname} • loading component",
-        "**/app/**/loading.tsx": "${dirname} • loading component",
+    "**/app/**/[[]*[]]/[[]*[]]/loading.tsx": "${dirname(2)}/${dirname(1)}/${dirname} • loading component",
+    "**/app/**/[[]*[]]/loading.tsx": "${dirname(1)}/${dirname} • loading component",
+    "**/app/**/loading.tsx": "${dirname} • loading component",
 
-        "**/src/**/route.ts": "${dirname(1)}/${dirname} • route",
-        "**/src/**/index.tsx": "${dirname} • component",
+    "**/src/**/route.ts": "${dirname(1)}/${dirname} • route",
+    "**/src/**/index.tsx": "${dirname} • component",
 
-        "**/src/database/repositories/*/index.ts": "${dirname} • db repository",
-        "**/src/database/models/*.ts": "${filename} • db model",
-        "**/src/database/schemas/*.ts": "${filename} • db schema",
+    "**/packages/database/src/repositories/*/index.ts": "${dirname} • db repository",
+    "**/packages/database/src/models/*.ts": "${filename} • db model",
+    "**/packages/database/src/schemas/*.ts": "${filename} • db schema",
 
-        "**/src/services/*.ts": "${filename} • service",
-        "**/src/services/*/client.ts": "${dirname} • client service",
-        "**/src/services/*/server.ts": "${dirname} • server service",
-        
-        "**/src/store/*/action.ts": "${dirname} • action",
-        "**/src/store/*/slices/*/action.ts": "${dirname(2)}/${dirname} • action",
-        "**/src/store/*/slices/*/actions/*.ts": "${dirname(1)}/${dirname}/${filename} • action",
-        
-        "**/src/store/*/initialState.ts": "${dirname} • state",
-        "**/src/store/*/slices/*/initialState.ts": "${dirname(2)}/${dirname} • state",
-        
-        "**/src/store/*/selectors.ts": "${dirname} • selectors",
-        "**/src/store/*/slices/*/selectors.ts": "${dirname(2)}/${dirname} • selectors",
-        
-        "**/src/store/*/reducer.ts": "${dirname} • reducer",
-        "**/src/store/*/slices/*/reducer.ts": "${dirname(2)}/${dirname} • reducer",
-        
-        "**/src/config/modelProviders/*.ts": "${filename} • provider",
-        "**/src/config/aiModels/*.ts": "${filename} • model",
-        "**/src/config/paramsSchemas/*/*.json": "${dirname(1)}/${filename} • params",
-        "**/packages/model-runtime/src/*/index.ts": "${dirname} • runtime",
-        
-        "**/src/server/services/*/index.ts": "${dirname} • server/service",
-        "**/src/server/routers/lambda/*.ts": "${filename} • lambda",
-        "**/src/server/routers/async/*.ts": "${filename} • async",
-        "**/src/server/routers/edge/*.ts": "${filename} • edge",
+    "**/src/services/*.ts": "${filename} • service",
+    "**/src/services/*/client.ts": "${dirname} • client service",
+    "**/src/services/*/server.ts": "${dirname} • server service",
 
-        "**/src/locales/default/*.ts": "${filename} • locale",
-    },
-    "eslint.validate": [
-        "javascript",
-        "javascriptreact",
-        "typescript",
-        "typescriptreact",
-        "markdown",
-        // support mdx
-        "mdx"
-    ]
-}
+    "**/src/store/*/action.ts": "${dirname} • action",
+    "**/src/store/*/slices/*/action.ts": "${dirname(2)}/${dirname} • action",
+    "**/src/store/*/slices/*/actions/*.ts": "${dirname(1)}/${dirname}/${filename} • action",
+
+    "**/src/store/*/initialState.ts": "${dirname} • state",
+    "**/src/store/*/slices/*/initialState.ts": "${dirname(2)}/${dirname} • state",
+
+    "**/src/store/*/selectors.ts": "${dirname} • selectors",
+    "**/src/store/*/slices/*/selectors.ts": "${dirname(2)}/${dirname} • selectors",
+
+    "**/src/store/*/reducer.ts": "${dirname} • reducer",
+    "**/src/store/*/slices/*/reducer.ts": "${dirname(2)}/${dirname} • reducer",
+
+    "**/src/config/modelProviders/*.ts": "${filename} • provider",
+    "**/packages/model-bank/src/aiModels/*.ts": "${filename} • model",
+    "**/packages/model-runtime/src/*/index.ts": "${dirname} • runtime",
+
+    "**/src/server/services/*/index.ts": "${dirname} • server/service",
+    "**/src/server/routers/lambda/*.ts": "${filename} • lambda",
+    "**/src/server/routers/async/*.ts": "${filename} • async",
+    "**/src/server/routers/edge/*.ts": "${filename} • edge",
+
+    "**/src/locales/default/*.ts": "${filename} • locale"
+  }
+}

AGENTS.md

@@ -0,0 +1,133 @@
+# LobeChat Development Guidelines
+
+This document serves as a comprehensive guide for all team members when developing LobeChat.
+
+## Tech Stack
+
+Built with modern technologies:
+
+- **Frontend**: Next.js 15, React 19, TypeScript
+- **UI Components**: Ant Design, @lobehub/ui, antd-style
+- **State Management**: Zustand, SWR
+- **Database**: PostgreSQL, PGLite, Drizzle ORM
+- **Testing**: Vitest, Testing Library
+- **Package Manager**: pnpm (monorepo structure)
+- **Build Tools**: Next.js (Turbopack in dev, Webpack in prod), Vitest
+
+## Directory Structure
+
+The project follows a well-organized monorepo structure:
+
+- `apps/` - Main applications
+- `packages/` - Shared packages and libraries
+- `src/` - Main source code
+- `docs/` - Documentation
+- `.cursor/rules/` - Development rules and guidelines
+
+## Development Workflow
+
+### Git Workflow
+
+- Use rebase for git pull: `git pull --rebase`
+- Git commit messages should prefix with gitmoji
+- Git branch name format: `username/feat/feature-name`
+- Use `.github/PULL_REQUEST_TEMPLATE.md` for PR descriptions
+
+### Package Management
+
+- Use `pnpm` as the primary package manager
+- Use `bun` to run npm scripts
+- Use `bunx` to run executable npm packages
+- Navigate to specific packages using `cd packages/<package-name>`
+
+### Code Style Guidelines
+
+#### TypeScript
+
+- Follow strict TypeScript practices for type safety and code quality
+- Use proper type annotations
+- Prefer interfaces over types for object shapes
+- Use generics for reusable components
+
+#### React Components
+
+- Use functional components with hooks
+- Follow the component structure guidelines
+- Use antd-style & @lobehub/ui for styling
+- Implement proper error boundaries
+
+#### Database Schema
+
+- Follow Drizzle ORM naming conventions
+- Use plural snake_case for table names
+- Implement proper foreign key relationships
+- Follow the schema style guide
+
+### Testing Strategy
+
+**Required Rule**: `testing-guide/testing-guide.mdc`
+
+**Commands**:
+
+- Web: `bunx vitest run --silent='passed-only' '[file-path-pattern]'`
+- Packages: `cd packages/[package-name] && bunx vitest run --silent='passed-only' '[file-path-pattern]'`
+
+**Important Notes**:
+
+- Wrap file paths in single quotes to avoid shell expansion
+- Never run `bun run test` - this runs all tests and takes ~10 minutes
+- If a test fails twice, stop and ask for help
+- Always add tests for new code
+
+### Type Checking
+
+- Use `bun run type-check` to check for type errors
+- Ensure all TypeScript errors are resolved before committing
+
+### Internationalization
+
+- Add new keys to `src/locales/default/namespace.ts`
+- Translate at least `zh-CN` files for development preview
+- Use hierarchical nested objects, not flat keys
+- Don't run `pnpm i18n` manually (handled by CI)
+
+## Available Development Rules
+
+The project provides comprehensive rules in `.cursor/rules/` directory:
+
+### Core Development
+
+- `backend-architecture.mdc` - Three-layer architecture and data flow
+- `react-component.mdc` - Component patterns and UI library usage
+- `drizzle-schema-style-guide.mdc` - Database schema conventions
+- `define-database-model.mdc` - Model templates and CRUD patterns
+- `i18n.mdc` - Internationalization workflow
+
+### State Management & UI
+
+- `zustand-slice-organization.mdc` - Store organization patterns
+- `zustand-action-patterns.mdc` - Action implementation patterns
+- `packages/react-layout-kit.mdc` - Flex layout component usage
+
+### Testing & Quality
+
+- `testing-guide/testing-guide.mdc` - Comprehensive testing strategy
+- `code-review.mdc` - Code review process and standards
+
+### Desktop (Electron)
+
+- `desktop-feature-implementation.mdc` - Main/renderer process patterns
+- `desktop-local-tools-implement.mdc` - Tool integration workflow
+- `desktop-menu-configuration.mdc` - Menu system configuration
+- `desktop-window-management.mdc` - Window management patterns
+- `desktop-controller-tests.mdc` - Controller testing guide
+
+## Best Practices
+
+- **Conservative for existing code, modern approaches for new features**
+- **Code Language**: Use Chinese for files with existing Chinese comments, American English for new files
+- Always add tests for new functionality
+- Follow the established patterns in the codebase
+- Use proper error handling and logging
+- Implement proper accessibility features
+- Consider internationalization from the start

CLAUDE.md

@@ -2,45 +2,30 @@
 
 This document serves as a shared guideline for all team members when using Claude Code in this repository.
 
-## Suggestions
+## Tech Stack
 
-- When searching the project source code, it is recommended to exclude: `src/database/migrations/meta`, `**/*.test.*`, `**/__snapshots__`, `**/fixtures`
-- Please store all temporary scripts (such as migration and refactoring scripts) in the `docs/.local/` directory; the contents of this folder will not be committed.
+read @.cursor/rules/project-introduce.mdc
 
-## Technologies Stack
+## Directory Structure
 
-read @.cursor/rules/project-introduce.mdc for more details.
-
-### Directory Structure
-
-```plaintext
-src/
-├── app/                 # Next.js App Router
-├── features/            # Feature-based UI components
-├── store/              # Zustand state stores
-├── services/           # Client services (tRPC/Model calls)
-├── server/             # Server-side (tRPC routers, services)
-├── database/           # Schemas, models, repositories
-├── libs/               # External library integrations
-```
-
-### Data Flow
-
-- **Client DB Version**: UI → Zustand → Service → Model → PGLite
-- **Server DB Version**: UI → Zustand → Service → tRPC → Repository/Model → PostgreSQL
+read @.cursor/rules/project-structure.mdc
 
 ## Development
 
 ### Git Workflow
 
-- use rebase for git pull.
-- git commit message should prefix with gitmoji.
+- use rebase for git pull
+- git commit message should prefix with gitmoji
 - git branch name format example: tj/feat/feature-name
 - use .github/PULL_REQUEST_TEMPLATE.md to generate pull request description
 
 ### Package Management
 
-this is a monorepo project and we use `pnpm` as package manager
+This repository adopts a monorepo structure.
+
+- Use `pnpm` as the primary package manager for dependency management
+- Use `bun` to run npm scripts
+- Use `bunx` to run executable npm packages
 
 ### TypeScript Code Style Guide
 
@@ -59,10 +44,13 @@ see @.cursor/rules/typescript.mdc
 Testing work follows the Rule-Aware Task Execution system above.
 
 - **Required Rule**: `testing-guide/testing-guide.mdc`
-- **Command**: `npx vitest run --config vitest.config.ts '[file-path-pattern]'`, wrapped in single quotes to avoid shell expansion
+- **Command**:
+  - web: `bunx vitest run --silent='passed-only' '[file-path-pattern]'`
+  - packages(eg: database): `cd packages/database && bunx vitest run --silent='passed-only' '[file-path-pattern]'`
 
 **Important**:
 
+- wrapped the file path in single quotes to avoid shell expansion
 - Never run `bun run test` etc to run tests, this will run all tests and cost about 10mins
 - If try to fix the same test twice, but still failed, stop and ask for help.
 
@@ -70,12 +58,11 @@ Testing work follows the Rule-Aware Task Execution system above.
 
 - use `bun run type-check` to check type errors.
 
-### Internationalization
+### i18n
 
 - **Keys**: Add to `src/locales/default/namespace.ts`
-- **Dev**: Translate at least `zh-CN` files for preview
-- **Structure**: Hierarchical nested objects, not flat keys
-- **Script**: DON'T run `pnpm i18n` (user/CI handles it)
+- **Dev**: Translate `locales/zh-CN/namespace.json` locale file only for preview
+- DON'T run `pnpm i18n`, let CI auto handle it
 
 ## Rules Index
 
@@ -91,12 +78,13 @@ Some useful rules of this project. Read them when needed.
 - `react-component.mdc` - antd-style, Lobe UI usage
 - `drizzle-schema-style-guide.mdc` - Schema naming, patterns
 - `define-database-model.mdc` - Model templates, CRUD patterns
+- `i18n.mdc` - Internationalization workflow
 
 **State & UI**
 
 - `zustand-slice-organization.mdc` - Store organization
 - `zustand-action-patterns.mdc` - Action patterns
-- `packages/react-layout-kit.mdc` - Layout components usage
+- `packages/react-layout-kit.mdc` - flex layout components usage
 
 **Testing & Quality**
 
@@ -110,8 +98,3 @@ Some useful rules of this project. Read them when needed.
 - `desktop-menu-configuration.mdc` - App menu, context menu, tray menu
 - `desktop-window-management.mdc` - Window creation, state management, multi-window
 - `desktop-controller-tests.mdc` - Controller unit testing guide
-
-**Development Tools**
-
-- `i18n.mdc` - Internationalization workflow
-- `debug.mdc` - Debugging strategies

Dockerfile

@@ -66,7 +66,7 @@ ENV NEXT_PUBLIC_ANALYTICS_UMAMI="${NEXT_PUBLIC_ANALYTICS_UMAMI}" \
     NEXT_PUBLIC_UMAMI_WEBSITE_ID="${NEXT_PUBLIC_UMAMI_WEBSITE_ID}"
 
 # Node
-ENV NODE_OPTIONS="--max-old-space-size=8192"
+ENV NODE_OPTIONS="--max-old-space-size=6144"
 
 WORKDIR /app
 
@@ -194,6 +194,8 @@ ENV \
     MODELSCOPE_API_KEY="" MODELSCOPE_MODEL_LIST="" MODELSCOPE_PROXY_URL="" \
     # Moonshot
     MOONSHOT_API_KEY="" MOONSHOT_MODEL_LIST="" MOONSHOT_PROXY_URL="" \
+    # Nebius
+    NEBIUS_API_KEY="" NEBIUS_MODEL_LIST="" NEBIUS_PROXY_URL="" \
     # Novita
     NOVITA_API_KEY="" NOVITA_MODEL_LIST="" \
     # Nvidia NIM
@@ -249,7 +251,11 @@ ENV \
     # Infini-AI
     INFINIAI_API_KEY="" INFINIAI_MODEL_LIST="" \
     # 302.AI
-    AI302_API_KEY="" AI302_MODEL_LIST=""
+    AI302_API_KEY="" AI302_MODEL_LIST="" \
+    # FAL
+    FAL_API_KEY="" FAL_MODEL_LIST="" \
+    # BFL
+    BFL_API_KEY="" BFL_MODEL_LIST=""
 
 USER nextjs
 

Dockerfile.database

@@ -74,7 +74,7 @@ ENV NEXT_PUBLIC_ANALYTICS_UMAMI="${NEXT_PUBLIC_ANALYTICS_UMAMI}" \
     NEXT_PUBLIC_UMAMI_WEBSITE_ID="${NEXT_PUBLIC_UMAMI_WEBSITE_ID}"
 
 # Node
-ENV NODE_OPTIONS="--max-old-space-size=8192"
+ENV NODE_OPTIONS="--max-old-space-size=6144"
 
 WORKDIR /app
 
@@ -120,7 +120,7 @@ COPY --from=base /distroless/ /
 COPY --from=builder /app/.next/standalone /app/
 
 # Copy database migrations
-COPY --from=builder /app/src/database/migrations /app/migrations
+COPY --from=builder /app/packages/database/migrations /app/migrations
 COPY --from=builder /app/scripts/migrateServerDB/docker.cjs /app/docker.cjs
 COPY --from=builder /app/scripts/migrateServerDB/errorHint.js /app/errorHint.js
 
@@ -236,6 +236,8 @@ ENV \
     MODELSCOPE_API_KEY="" MODELSCOPE_MODEL_LIST="" MODELSCOPE_PROXY_URL="" \
     # Moonshot
     MOONSHOT_API_KEY="" MOONSHOT_MODEL_LIST="" MOONSHOT_PROXY_URL="" \
+    # Nebius
+    NEBIUS_API_KEY="" NEBIUS_MODEL_LIST="" NEBIUS_PROXY_URL="" \
     # Novita
     NOVITA_API_KEY="" NOVITA_MODEL_LIST="" \
     # Nvidia NIM
@@ -291,7 +293,11 @@ ENV \
     # Infini-AI
     INFINIAI_API_KEY="" INFINIAI_MODEL_LIST="" \
     # 302.AI
-    AI302_API_KEY="" AI302_MODEL_LIST=""
+    AI302_API_KEY="" AI302_MODEL_LIST="" \
+    # FAL
+    FAL_API_KEY="" FAL_MODEL_LIST="" \
+    # BFL
+    BFL_API_KEY="" BFL_MODEL_LIST=""
 
 USER nextjs
 

Dockerfile.pglite

@@ -68,7 +68,7 @@ ENV NEXT_PUBLIC_ANALYTICS_UMAMI="${NEXT_PUBLIC_ANALYTICS_UMAMI}" \
     NEXT_PUBLIC_UMAMI_WEBSITE_ID="${NEXT_PUBLIC_UMAMI_WEBSITE_ID}"
 
 # Node
-ENV NODE_OPTIONS="--max-old-space-size=8192"
+ENV NODE_OPTIONS="--max-old-space-size=6144"
 
 WORKDIR /app
 
@@ -196,6 +196,8 @@ ENV \
     MODELSCOPE_API_KEY="" MODELSCOPE_MODEL_LIST="" MODELSCOPE_PROXY_URL="" \
     # Moonshot
     MOONSHOT_API_KEY="" MOONSHOT_MODEL_LIST="" MOONSHOT_PROXY_URL="" \
+    # Nebius
+    NEBIUS_API_KEY="" NEBIUS_MODEL_LIST="" NEBIUS_PROXY_URL="" \
     # Novita
     NOVITA_API_KEY="" NOVITA_MODEL_LIST="" \
     # Nvidia NIM
@@ -247,7 +249,11 @@ ENV \
     # Infini-AI
     INFINIAI_API_KEY="" INFINIAI_MODEL_LIST="" \
     # 302.AI
-    AI302_API_KEY="" AI302_MODEL_LIST=""
+    AI302_API_KEY="" AI302_MODEL_LIST="" \
+    # FAL
+    FAL_API_KEY="" FAL_MODEL_LIST="" \
+    # BFL
+    BFL_API_KEY="" BFL_MODEL_LIST=""
 
 USER nextjs
 

README.md

@@ -150,7 +150,7 @@ From productivity tools to development environments, discover new ways to extend
 
 **Peak Performance, Zero Distractions**
 
-Get the full LobeChat experience without browser limitations—lightweight, focused, and always ready to go. Our desktop application provides a dedicated environment for your AI interactions, ensuring optimal performance and minimal distractions.
+Get the full LobeChat experience without browser limitations—comprehensive, focused, and always ready to go. Our desktop application provides a dedicated environment for your AI interactions, ensuring optimal performance and minimal distractions.
 
 Experience faster response times, better resource management, and a more stable connection to your AI assistant. The desktop app is designed for users who demand the best performance from their AI tools.
 
@@ -481,7 +481,7 @@ We deeply understand the importance of providing a seamless experience for users
 Therefore, we have adopted Progressive Web Application ([PWA](https://support.google.com/chrome/answer/9658361)) technology,
 a modern web technology that elevates web applications to an experience close to that of native apps.
 
-Through PWA, LobeChat can offer a highly optimized user experience on both desktop and mobile devices while maintaining its lightweight and high-performance characteristics.
+Through PWA, LobeChat can offer a highly optimized user experience on both desktop and mobile devices while maintaining high-performance characteristics.
 Visually and in terms of feel, we have also meticulously designed the interface to ensure it is indistinguishable from native apps,
 providing smooth animations, responsive layouts, and adapting to different device screen resolutions.
 

apps/desktop/electron-builder.js

@@ -64,7 +64,7 @@ const config = {
   linux: {
     category: 'Utility',
     maintainer: 'electronjs.org',
-    target: ['AppImage', 'snap', 'deb'],
+    target: ['AppImage', 'snap', 'deb', 'rpm', 'tar.gz'],
   },
   mac: {
     compression: 'maximum',

apps/desktop/package.json

@@ -31,6 +31,7 @@
   "dependencies": {
     "electron-updater": "^6.6.2",
     "electron-window-state": "^5.0.3",
+    "fetch-socks": "^1.3.2",
     "get-port-please": "^3.1.2",
     "pdfjs-dist": "4.10.38"
   },
@@ -51,7 +52,7 @@
     "@typescript/native-preview": "7.0.0-dev.20250711.1",
     "consola": "^3.1.0",
     "cookie": "^1.0.2",
-    "electron": "^37.2.0",
+    "electron": "^37.4.0",
     "electron-builder": "^26.0.12",
     "electron-is": "^3.0.0",
     "electron-log": "^5.3.3",

apps/desktop/src/main/const/store.ts

@@ -25,6 +25,7 @@ export const defaultProxySettings: NetworkProxySettings = {
  * 存储默认值
  */
 export const STORE_DEFAULTS: ElectronMainStore = {
+  autoUpdateNotificationEnabled: true,
   dataSyncConfig: { storageMode: 'local' },
   encryptedTokens: {},
   locale: 'auto',

apps/desktop/src/main/controllers/DownloadCtr.ts

@@ -0,0 +1,41 @@
+import { createLogger } from '@/utils/logger';
+
+import { ControllerModule, ipcClientEvent } from './index';
+
+// Create logger
+const logger = createLogger('controllers:DownloadCtr');
+
+/**
+ * 下载控制器
+ * 处理桌面应用的下载相关功能，包括自动更新通知设置
+ */
+export default class DownloadCtr extends ControllerModule {
+  /**
+   * 获取自动更新通知设置
+   */
+  @ipcClientEvent('getAutoUpdateNotificationEnabled')
+  async getAutoUpdateNotificationEnabled(): Promise<boolean> {
+    try {
+      const enabled = this.app.storeManager.get('autoUpdateNotificationEnabled', true);
+      logger.debug('Retrieved auto update notification setting:', enabled);
+      return enabled;
+    } catch (error) {
+      logger.error('Failed to get auto update notification setting:', error);
+      return true; // 默认启用
+    }
+  }
+
+  /**
+   * 设置自动更新通知
+   */
+  @ipcClientEvent('setAutoUpdateNotificationEnabled')
+  async setAutoUpdateNotificationEnabled(enabled: boolean): Promise<void> {
+    try {
+      this.app.storeManager.set('autoUpdateNotificationEnabled', enabled);
+      logger.info('Auto update notification setting updated:', enabled);
+    } catch (error) {
+      logger.error('Failed to set auto update notification setting:', error);
+      throw error;
+    }
+  }
+}

apps/desktop/src/main/modules/networkProxy/dispatcher.ts

@@ -1,4 +1,5 @@
 import { NetworkProxySettings } from '@lobechat/electron-client-ipc';
+import { SocksProxies, socksDispatcher } from 'fetch-socks';
 import { Agent, ProxyAgent, getGlobalDispatcher, setGlobalDispatcher } from 'undici';
 
 import { createLogger } from '@/utils/logger';
@@ -91,8 +92,29 @@ export class ProxyDispatcherManager {
    */
   static createProxyAgent(proxyType: string, proxyUrl: string) {
     try {
-      // undici 的 ProxyAgent 支持 http, https 和 socks5
-      return new ProxyAgent({ uri: proxyUrl });
+      if (proxyType === 'socks5') {
+        // 解析 SOCKS5 代理 URL
+        const url = new URL(proxyUrl);
+        const socksProxies: SocksProxies = [
+          {
+            host: url.hostname,
+            port: parseInt(url.port, 10),
+            type: 5,
+            ...(url.username && url.password
+              ? {
+                  password: url.password,
+                  userId: url.username,
+                }
+              : {}),
+          },
+        ];
+
+        // 使用 fetch-socks 处理 SOCKS5 代理
+        return socksDispatcher(socksProxies);
+      } else {
+        // undici 的 ProxyAgent 支持 http, https
+        return new ProxyAgent({ uri: proxyUrl });
+      }
     } catch (error) {
       logger.error(`Failed to create proxy agent for ${proxyType}:`, error);
       throw new Error(

apps/desktop/src/main/types/store.ts

@@ -1,6 +1,7 @@
 import { DataSyncConfig, NetworkProxySettings } from '@lobechat/electron-client-ipc';
 
 export interface ElectronMainStore {
+  autoUpdateNotificationEnabled: boolean;
   dataSyncConfig: DataSyncConfig;
   encryptedTokens: {
     accessToken?: string;

changelog/v1.json

@@ -1,4 +1,338 @@
 [
+  {
+    "children": {
+      "fixes": ["Not use branch topic when this topic is not save."]
+    },
+    "date": "2025-09-05",
+    "version": "1.123.2"
+  },
+  {
+    "children": {
+      "improvements": ["Update i18n."]
+    },
+    "date": "2025-09-05",
+    "version": "1.123.1"
+  },
+  {
+    "children": {
+      "features": ["Add NewAPI as a router provider for multi-model aggregation."]
+    },
+    "date": "2025-09-04",
+    "version": "1.123.0"
+  },
+  {
+    "children": {
+      "improvements": ["Make LobeNextAuthDBAdapter Edge Compatible."]
+    },
+    "date": "2025-09-04",
+    "version": "1.122.7"
+  },
+  {
+    "children": {},
+    "date": "2025-09-04",
+    "version": "1.122.6"
+  },
+  {
+    "children": {
+      "improvements": ["Make LobeNextAuthDBAdapter Edge Compatible."]
+    },
+    "date": "2025-09-04",
+    "version": "1.122.5"
+  },
+  {
+    "children": {
+      "improvements": ["Update i18n."]
+    },
+    "date": "2025-09-04",
+    "version": "1.122.4"
+  },
+  {
+    "children": {
+      "fixes": ["Support base64 image from markdown image syntax."],
+      "improvements": ["Update the price of the o3 model in OpenRouter."]
+    },
+    "date": "2025-09-04",
+    "version": "1.122.3"
+  },
+  {
+    "children": {},
+    "date": "2025-09-04",
+    "version": "1.122.2"
+  },
+  {
+    "children": {},
+    "date": "2025-09-04",
+    "version": "1.122.1"
+  },
+  {
+    "children": {
+      "features": ["Refactor to speed up send message in server mode."]
+    },
+    "date": "2025-09-04",
+    "version": "1.122.0"
+  },
+  {
+    "children": {
+      "fixes": ["Fix socks5 proxy not work problem, fix virtuaso minheight was null."]
+    },
+    "date": "2025-09-03",
+    "version": "1.121.1"
+  },
+  {
+    "children": {
+      "features": ["Add nano banana Chinese prompt notify."]
+    },
+    "date": "2025-09-03",
+    "version": "1.121.0"
+  },
+  {
+    "children": {
+      "improvements": ["Update i18n."]
+    },
+    "date": "2025-09-02",
+    "version": "1.120.7"
+  },
+  {
+    "children": {
+      "improvements": ["Add upload hint for non-visual model."]
+    },
+    "date": "2025-09-01",
+    "version": "1.120.6"
+  },
+  {
+    "children": {},
+    "date": "2025-09-01",
+    "version": "1.120.5"
+  },
+  {
+    "children": {
+      "improvements": ["Adjust ControlsForm component to adapt to mobile phone display."]
+    },
+    "date": "2025-09-01",
+    "version": "1.120.4"
+  },
+  {
+    "children": {
+      "improvements": ["Support new provider Nebius."]
+    },
+    "date": "2025-09-01",
+    "version": "1.120.3"
+  },
+  {
+    "children": {
+      "improvements": ["Remove base path."]
+    },
+    "date": "2025-08-31",
+    "version": "1.120.2"
+  },
+  {
+    "children": {
+      "improvements": ["Update i18n."]
+    },
+    "date": "2025-08-31",
+    "version": "1.120.1"
+  },
+  {
+    "children": {
+      "improvements": ["Remove webrtc sync feature flag."],
+      "features": ["Rename Gemini 2.5 flash image to Nano Banana."]
+    },
+    "date": "2025-08-30",
+    "version": "1.120.0"
+  },
+  {
+    "children": {},
+    "date": "2025-08-30",
+    "version": "1.119.2"
+  },
+  {
+    "children": {
+      "fixes": ["Update enableStreaming name."]
+    },
+    "date": "2025-08-30",
+    "version": "1.119.1"
+  },
+  {
+    "children": {
+      "features": ["Added support for Azure OpenAI Image Generation."]
+    },
+    "date": "2025-08-30",
+    "version": "1.119.0"
+  },
+  {
+    "children": {
+      "improvements": ["Update DeepSeek V3.1 & Gemini 2.5 Flash Image Preview models."]
+    },
+    "date": "2025-08-30",
+    "version": "1.118.8"
+  },
+  {
+    "children": {
+      "improvements": ["Update i18n."]
+    },
+    "date": "2025-08-30",
+    "version": "1.118.7"
+  },
+  {
+    "children": {
+      "improvements": ["Support non-stream mode."]
+    },
+    "date": "2025-08-29",
+    "version": "1.118.6"
+  },
+  {
+    "children": {
+      "improvements": ["Fix clerk scrollBox style, ModelFetcher support getting prices."]
+    },
+    "date": "2025-08-29",
+    "version": "1.118.5"
+  },
+  {
+    "children": {
+      "improvements": ["Fix chat session part switch theme issue."]
+    },
+    "date": "2025-08-29",
+    "version": "1.118.4"
+  },
+  {
+    "children": {
+      "fixes": ["Correct totalOutputTokens calculation for XAI provider."]
+    },
+    "date": "2025-08-29",
+    "version": "1.118.3"
+  },
+  {
+    "children": {
+      "improvements": ["Add Grok Code Fast 1 model."]
+    },
+    "date": "2025-08-29",
+    "version": "1.118.2"
+  },
+  {
+    "children": {
+      "fixes": ["Add Content-Security-Policy env."],
+      "improvements": ["Support Gemini URL context tool."]
+    },
+    "date": "2025-08-29",
+    "version": "1.118.1"
+  },
+  {
+    "children": {
+      "features": ["Add new provider AkashChat."]
+    },
+    "date": "2025-08-29",
+    "version": "1.118.0"
+  },
+  {
+    "children": {
+      "improvements": ["Move chat item into chat."]
+    },
+    "date": "2025-08-29",
+    "version": "1.117.1"
+  },
+  {
+    "children": {
+      "features": [
+        "Ai image support Gemini 2.5 Flash Image, Support Gemini 2.5 Flash Image Preview in OpenRouter."
+      ],
+      "improvements": ["Update i18n."]
+    },
+    "date": "2025-08-29",
+    "version": "1.117.0"
+  },
+  {
+    "children": {
+      "improvements": ["Support html preview."]
+    },
+    "date": "2025-08-28",
+    "version": "1.116.4"
+  },
+  {
+    "children": {
+      "fixes": ["Fix desktop route error."]
+    },
+    "date": "2025-08-28",
+    "version": "1.116.3"
+  },
+  {
+    "children": {},
+    "date": "2025-08-28",
+    "version": "1.116.2"
+  },
+  {
+    "children": {},
+    "date": "2025-08-27",
+    "version": "1.116.1"
+  },
+  {
+    "children": {
+      "features": ["Add gemini 2.5 flash image for vertex ai."]
+    },
+    "date": "2025-08-27",
+    "version": "1.116.0"
+  },
+  {
+    "children": {},
+    "date": "2025-08-26",
+    "version": "1.115.0"
+  },
+  {
+    "children": {},
+    "date": "2025-08-22",
+    "version": "1.114.6"
+  },
+  {
+    "children": {
+      "improvements": ["Update mistral model vision ability."]
+    },
+    "date": "2025-08-22",
+    "version": "1.114.5"
+  },
+  {
+    "children": {
+      "improvements": ["Move database to packages."]
+    },
+    "date": "2025-08-22",
+    "version": "1.114.4"
+  },
+  {
+    "children": {},
+    "date": "2025-08-21",
+    "version": "1.114.3"
+  },
+  {
+    "children": {
+      "fixes": ["Can't load custom provider config."]
+    },
+    "date": "2025-08-21",
+    "version": "1.114.2"
+  },
+  {
+    "children": {
+      "improvements": ["Move chain into @lobechat/prompts."]
+    },
+    "date": "2025-08-21",
+    "version": "1.114.1"
+  },
+  {
+    "children": {},
+    "date": "2025-08-19",
+    "version": "1.114.0"
+  },
+  {
+    "children": {
+      "fixes": [
+        "Support Grok thinking models in AiHubMix, The 'stream_options' parameter is only allowed when 'stream' is enabled."
+      ]
+    },
+    "date": "2025-08-19",
+    "version": "1.113.3"
+  },
+  {
+    "children": {},
+    "date": "2025-08-18",
+    "version": "1.113.2"
+  },
   {
     "children": {},
     "date": "2025-08-17",

codecov.yml

@@ -1,11 +1,39 @@
+component_management:
+  individual_components:
+    # App architecture layers
+    - component_id: app_store
+      name: "Store"
+      paths:
+        - src/store/**
+    - component_id: app_services
+      name: "Services"
+      paths:
+        - src/services/**
+    - component_id: app_server
+      name: "Server"
+      paths:
+        - src/server/**
+    - component_id: app_libs
+      name: "Libs"
+      paths:
+        - src/libs/**
+    - component_id: app_utils
+      name: "Utils"
+      paths:
+        - src/utils/**
+
 coverage:
   status:
     project:
       default: off
-      server:
+      database:
         flags:
-          - server
+          - database
       app:
         flags:
           - app
     patch: off
+
+
+comment:
+  layout: "header, diff, flags, components"  # show component info in the PR comment

docker-compose.development.yml

@@ -0,0 +1,40 @@
+name: lobe-chat-development
+services:
+  network-service:
+    image: alpine
+    container_name: lobe-network
+    restart: always
+    ports:
+      - '${MINIO_PORT}:${MINIO_PORT}' # MinIO API
+      - '9001:9001' # MinIO Console
+      - '${CASDOOR_PORT}:${CASDOOR_PORT}' # Casdoor
+    command: tail -f /dev/null
+    networks:
+      - lobe-network
+
+  postgresql:
+    extends:
+      file: docker-compose/local/docker-compose.yml
+      service: postgresql
+  minio:
+    extends:
+      file: docker-compose/local/docker-compose.yml
+      service: minio
+  casdoor:
+    extends:
+      file: docker-compose/local/docker-compose.yml
+      service: casdoor
+  searxng:
+    extends:
+      file: docker-compose/local/docker-compose.yml
+      service: searxng
+
+volumes:
+  data:
+    driver: local
+  s3_data:
+    driver: local
+
+networks:
+  lobe-network:
+    driver: bridge

docker-compose/local/.gitignore

@@ -0,0 +1,2 @@
+data/
+s3_data/

docs/development/basic/feature-development.mdx

@@ -456,7 +456,7 @@ The project uses vitest for unit testing.
 
 Since our two new configuration fields are both optional, theoretically you could pass the tests without updating them. However, since we added the `openingQuestions` field to the `DEFAULT_AGENT_CONFIG` mentioned earlier, this causes many tests to calculate configurations that include this field, so we still need to update some test snapshots.
 
-For the current scenario, I recommend running the tests locally to see which tests fail, and then update them as needed. For example, for the test file `src/store/agent/slices/chat/selectors/agent.test.ts`, you need to run `npx vitest -u src/store/agent/slices/chat/selectors/agent.test.ts` to update the snapshot.
+For the current scenario, I recommend running the tests locally to see which tests fail, and then update them as needed. For example, for the test file `src/store/agent/slices/chat/selectors/agent.test.ts`, you need to run `bunx vitest -u src/store/agent/slices/chat/selectors/agent.test.ts` to update the snapshot.
 
 ## Summary
 

docs/development/basic/feature-development.zh-CN.mdx

@@ -456,7 +456,7 @@ export default WelcomeMessage;
 
 由于我们目前两个新的配置字段都是可选的，所以理论上你不更新测试也能跑通，不过由于我们把前面提到的默认配置 `DEFAULT_AGENT_CONFIG` 增加了 `openingQuestions` 字段，这导致很多测试计算出的配置都是有这个字段的，因此我们还是需要更新一部分测试的快照。
 
-对于当前这个场景，我建议是本地直接跑下测试，看哪些测试失败了，针对需要更新，例如测试文件 `src/store/agent/slices/chat/selectors/agent.test.ts` 需要执行一下 `npx vitest -u src/store/agent/slices/chat/selectors/agent.test.ts` 更新快照。
+对于当前这个场景，我建议是本地直接跑下测试，看哪些测试失败了，针对需要更新，例如测试文件 `src/store/agent/slices/chat/selectors/agent.test.ts` 需要执行一下 `bunx vitest -u src/store/agent/slices/chat/selectors/agent.test.ts` 更新快照。
 
 ## 总结
 

docs/development/basic/image-generation-setup.mdx

@@ -0,0 +1,190 @@
+---
+title: Image Generation Development Setup
+description: Configure local environment for developing text-to-image and image-to-image features
+---
+
+# Image Generation Development Setup
+
+This guide helps developers set up the local environment for developing image generation features (text-to-image, image-to-image) with file storage capabilities.
+
+## Prerequisites
+
+- Docker installed and running
+- Node.js and pnpm installed
+- PostgreSQL client tools (optional, for debugging)
+
+<Callout type="warning">
+  **Security Notice**: This setup is designed for local development only. It uses default credentials and open permissions that are NOT suitable for production environments.
+</Callout>
+
+## Quick Setup
+
+Run the provided script to automatically set up all required services:
+
+```bash
+# Set up PostgreSQL and MinIO for image storage
+./scripts/setup-image-generation-dev.sh
+
+# Start the development server
+pnpm dev:desktop
+```
+
+This script will:
+
+1. Start PostgreSQL (no authentication for local development)
+2. Run database migrations to initialize schema
+3. Start MinIO (S3-compatible storage)
+4. Create and configure the storage bucket
+5. Add necessary S3 environment variables to `.env.desktop`
+
+## Architecture Overview
+
+The image generation feature requires:
+
+- **PostgreSQL**: Stores metadata about generated images
+- **MinIO/S3**: Stores the actual image files
+- **Server Mode**: Required for file handling (`NEXT_PUBLIC_SERVICE_MODE=server`)
+
+## Environment Configuration
+
+The following environment variables are automatically configured by the setup script:
+
+```bash
+# S3 Storage Configuration (MinIO for local development)
+S3_ACCESS_KEY_ID=minioadmin
+S3_SECRET_ACCESS_KEY=minioadmin
+S3_ENDPOINT=http://localhost:9000
+S3_BUCKET=lobe-chat
+S3_REGION=us-east-1
+S3_PUBLIC_DOMAIN=http://localhost:9000/lobe-chat
+S3_ENABLE_PATH_STYLE=1  # Required for MinIO
+```
+
+## Development Workflow
+
+### 1. Image Generation API
+
+When developing image generation features, generated images will be:
+
+1. Created by the AI model
+2. Uploaded to S3/MinIO via presigned URLs
+3. Metadata stored in PostgreSQL
+4. Served via the public S3 URL
+
+### 2. File Storage Structure
+
+```
+lobe-chat/           # S3 Bucket
+├── generated/       # Generated images
+│   └── {userId}/
+│       └── {sessionId}/
+│           └── {imageId}.png
+└── uploads/         # User uploads for image-to-image
+    └── {userId}/
+        └── {fileId}.{ext}
+```
+
+### 3. Testing Your Implementation
+
+After setting up the environment, you can test:
+
+```typescript
+// Example: Upload generated image
+const uploadUrl = await trpc.upload.createPresignedUrl.mutate({
+  filename: 'generated-image.png',
+  contentType: 'image/png',
+});
+
+// Upload to S3
+await fetch(uploadUrl, {
+  method: 'PUT',
+  body: imageBlob,
+  headers: { 'Content-Type': 'image/png' },
+});
+```
+
+## Manual Setup
+
+If you prefer to set up services manually:
+
+### PostgreSQL
+
+```bash
+docker run -d --name lobe-postgres \
+  -p 5432:5432 \
+  -e POSTGRES_HOST_AUTH_METHOD=trust \
+  -e POSTGRES_DB=postgres \
+  postgres:15
+```
+
+### MinIO
+
+```bash
+# Start MinIO
+docker run -d --name lobe-minio \
+  -p 9000:9000 -p 9001:9001 \
+  -e MINIO_ROOT_USER=minioadmin \
+  -e MINIO_ROOT_PASSWORD=minioadmin \
+  quay.io/minio/minio:RELEASE.2025-04-22T22-12-26Z \
+  server /data --console-address ":9001"
+
+# Create bucket
+docker run --rm \
+  --link lobe-minio:minio \
+  --entrypoint bash \
+  quay.io/minio/mc:RELEASE.2025-04-18T16-45-00Z \
+  -c "
+    mc config host add minio http://minio:9000 minioadmin minioadmin &&
+    mc mb minio/lobe-chat &&
+    mc anonymous set public minio/lobe-chat
+  "
+```
+
+## Service URLs
+
+- **PostgreSQL**: `postgres://postgres@localhost:5432/postgres`
+- **MinIO API**: `http://localhost:9000`
+- **MinIO Console**: `http://localhost:9001` (minioadmin/minioadmin)
+- **Application**: `http://localhost:3015`
+
+## Troubleshooting
+
+### Port Conflicts
+
+If ports are already in use:
+
+```bash
+# Check what's using the ports
+lsof -i :5432  # PostgreSQL
+lsof -i :9000  # MinIO API
+lsof -i :9001  # MinIO Console
+```
+
+### Reset Environment
+
+To completely reset your development environment:
+
+```bash
+# Stop and remove containers
+docker stop lobe-postgres lobe-minio
+docker rm lobe-postgres lobe-minio
+
+# Re-run setup
+./scripts/setup-image-generation-dev.sh
+```
+
+### Database Migrations
+
+The setup script runs migrations automatically. If you need to run them manually:
+
+```bash
+pnpm db:migrate
+```
+
+Note: In development mode with `pnpm dev:desktop`, migrations also run automatically on startup.
+
+## Related Documentation
+
+- [Server Database Setup](/docs/self-hosting/server-database)
+- [S3 Storage Configuration](/docs/self-hosting/advanced/s3)
+- [Environment Variables](/docs/self-hosting/environment-variables)

docs/development/basic/image-generation-setup.zh-CN.mdx

@@ -0,0 +1,190 @@
+---
+title: 图像生成开发环境配置
+description: 配置本地环境以开发文本生图和图像处理功能
+---
+
+# 图像生成开发环境配置
+
+本指南帮助开发者配置本地环境，用于开发图像生成功能（文生图、图生图）等和文件存储能力。
+
+## 前置条件
+
+- 已安装并运行 Docker
+- 已安装 Node.js 和 pnpm
+- PostgreSQL 客户端工具（可选，用于调试）
+
+<Callout type="warning">
+  **安全提醒**：此配置仅适用于本地开发。使用的默认凭据和开放权限不适合生产环境。
+</Callout>
+
+## 快速配置
+
+运行提供的脚本来自动配置所有必需的服务：
+
+```bash
+# 配置 PostgreSQL 和 MinIO 用于图像存储
+./scripts/setup-image-generation-dev.sh
+
+# 启动开发服务器
+pnpm dev:desktop
+```
+
+此脚本将执行：
+
+1. 启动 PostgreSQL（本地开发无需身份验证）
+2. 运行数据库迁移以初始化模式
+3. 启动 MinIO（S3 兼容存储）
+4. 创建并配置存储桶
+5. 在 `.env.desktop` 中添加必要的 S3 环境变量
+
+## 架构概览
+
+图像生成功能需要：
+
+- **PostgreSQL**：存储生成图像的元数据
+- **MinIO/S3**：存储实际的图像文件
+- **服务器模式**：文件处理所需（`NEXT_PUBLIC_SERVICE_MODE=server`）
+
+## 环境配置
+
+以下环境变量会被配置脚本自动设置：
+
+```bash
+# S3 存储配置（本地开发使用 MinIO）
+S3_ACCESS_KEY_ID=minioadmin
+S3_SECRET_ACCESS_KEY=minioadmin
+S3_ENDPOINT=http://localhost:9000
+S3_BUCKET=lobe-chat
+S3_REGION=us-east-1
+S3_PUBLIC_DOMAIN=http://localhost:9000/lobe-chat
+S3_ENABLE_PATH_STYLE=1  # MinIO 必需
+```
+
+## 开发工作流
+
+### 1. 图像生成 API
+
+在开发图像生成功能时，生成的图像将：
+
+1. 由 AI 模型创建
+2. 通过预签名 URL 上传到 S3/MinIO
+3. 元数据存储在 PostgreSQL 中
+4. 通过公共 S3 URL 提供服务
+
+### 2. 文件存储结构
+
+```
+lobe-chat/           # S3 存储桶
+├── generated/       # 生成的图像
+│   └── {userId}/
+│       └── {sessionId}/
+│           └── {imageId}.png
+└── uploads/         # 用户上传的图像处理文件
+    └── {userId}/
+        └── {fileId}.{ext}
+```
+
+### 3. 测试您的实现
+
+配置环境后，您可以测试：
+
+```typescript
+// 示例：上传生成的图像
+const uploadUrl = await trpc.upload.createPresignedUrl.mutate({
+  filename: 'generated-image.png',
+  contentType: 'image/png',
+});
+
+// 上传到 S3
+await fetch(uploadUrl, {
+  method: 'PUT',
+  body: imageBlob,
+  headers: { 'Content-Type': 'image/png' },
+});
+```
+
+## 手动配置
+
+如果您希望手动配置服务：
+
+### PostgreSQL
+
+```bash
+docker run -d --name lobe-postgres \
+  -p 5432:5432 \
+  -e POSTGRES_HOST_AUTH_METHOD=trust \
+  -e POSTGRES_DB=postgres \
+  postgres:15
+```
+
+### MinIO
+
+```bash
+# 启动 MinIO
+docker run -d --name lobe-minio \
+  -p 9000:9000 -p 9001:9001 \
+  -e MINIO_ROOT_USER=minioadmin \
+  -e MINIO_ROOT_PASSWORD=minioadmin \
+  quay.io/minio/minio:RELEASE.2025-04-22T22-12-26Z \
+  server /data --console-address ":9001"
+
+# 创建存储桶
+docker run --rm \
+  --link lobe-minio:minio \
+  --entrypoint bash \
+  quay.io/minio/mc:RELEASE.2025-04-18T16-45-00Z \
+  -c "
+    mc config host add minio http://minio:9000 minioadmin minioadmin &&
+    mc mb minio/lobe-chat &&
+    mc anonymous set public minio/lobe-chat
+  "
+```
+
+## 服务地址
+
+- **PostgreSQL**：`postgres://postgres@localhost:5432/postgres`
+- **MinIO API**：`http://localhost:9000`
+- **MinIO 控制台**：`http://localhost:9001` (minioadmin/minioadmin)
+- **应用程序**：`http://localhost:3015`
+
+## 故障排除
+
+### 端口冲突
+
+如果端口已被占用：
+
+```bash
+# 检查端口使用情况
+lsof -i :5432  # PostgreSQL
+lsof -i :9000  # MinIO API
+lsof -i :9001  # MinIO 控制台
+```
+
+### 重置环境
+
+要完全重置开发环境：
+
+```bash
+# 停止并删除容器
+docker stop lobe-postgres lobe-minio
+docker rm lobe-postgres lobe-minio
+
+# 重新运行配置
+./scripts/setup-image-generation-dev.sh
+```
+
+### 数据库迁移
+
+配置脚本会自动运行迁移。如需手动运行：
+
+```bash
+pnpm db:migrate
+```
+
+注意：在使用 `pnpm dev:desktop` 的开发模式下，迁移也会在启动时自动运行。
+
+## 相关文档
+
+- [服务器数据库配置](/docs/self-hosting/server-database)
+- [S3 存储配置](/docs/self-hosting/advanced/s3)
+- [环境变量](/docs/self-hosting/environment-variables)

docs/development/basic/setup-development.mdx

@@ -17,10 +17,14 @@ Before starting development on LobeChat, you need to install and configure some
 First, you need to install the following software:
 
 - Node.js: LobeChat is built on Node.js, so you need to install Node.js. We recommend installing the latest stable version.
-- Yarn: We use Yarn as the preferred package manager. You can download and install it from the Yarn official website.
-- PNPM: We use PNPM as an auxiliary package manager. You can download and install it from the PNPM official website.
+- PNPM: We use PNPM as the preferred package manager. You can download and install it from the [PNPM official website](https://pnpm.io/installation).
+- Bun: We use Bun as the npm scripts runner. You can download and install it from the [Bun official website](https://bun.com/docs/installation).
 - Git: We use Git for version control. You can download and install it from the Git official website.
-- IDE: You can choose your preferred integrated development environment (IDE). We recommend using WebStorm, a powerful IDE particularly suitable for TypeScript development.
+- IDE: You can choose your preferred integrated development environment (IDE). We recommend using WebStorm/VSCode.
+
+### VSCode Users
+
+We recommend installing the extensions listed in [.vscode/extensions.json](https://github.com/lobehub/lobe-chat/blob/main/.vscode/extensions.json) for the best development experience.
 
 ### Project Setup
 
@@ -32,24 +36,17 @@ After installing the above software, you can start setting up the LobeChat proje
 git clone https://github.com/lobehub/lobe-chat.git
 ```
 
-2. **Install dependencies**: Then, navigate to the project directory and use Yarn to install the project's dependencies:
+2. **Install dependencies**: Then, navigate to the project directory and use PNPM to install the project's dependencies:
 
 ```bash
 cd lobe-chat
-yarn install
-```
-
-If you are using PNPM, you can execute:
-
-```bash
-cd lobe-chat
-pnpm install
+pnpm i
 ```
 
 3. **Start the development server**: After installing the dependencies, you can start the development server:
 
 ```bash
-yarn run dev
+bun run dev
 ```
 
 Now, you can open `http://localhost:3010` in your browser, and you should see the welcome page of LobeChat. This indicates that you have successfully set up the development environment.

docs/development/basic/setup-development.zh-CN.mdx

@@ -17,10 +17,14 @@
 首先，你需要安装以下软件：
 
 - Node.js：LobeChat 是基于 Node.js 构建的，因此你需要安装 Node.js。我们建议安装最新的稳定版。
-- Bun：我们使用 Bun 作为首选包管理器。你可以从 Bun 的官方网站上下载并安装。
-- PNPM：我们使用 PNPM 作为辅助包管理器。你可以从 pnpm 的官方网站上下载并安装。
+- PNPM：我们使用 PNPM 作为管理器。你可以从 [pnpm 的官方网站](https://pnpm.io/installation) 上下载并安装。
+- Bun：我们使用 Bun 作为 npm scripts runner, 你可以从 [Bun 的官方网站](https://bun.com/docs/installation) 上下载并安装。
 - Git：我们使用 Git 进行版本控制。你可以从 Git 的官方网站上下载并安装。
-- IDE：你可以选择你喜欢的集成开发环境（IDE）。我们推荐使用 WebStorm，它是一款功能强大的 IDE，特别适合 TypeScript 开发。
+- IDE：你可以选择你喜欢的集成开发环境（IDE），我们推荐使用 WebStorm/VSCode。
+
+### VSCode 用户
+
+推荐安装 [.vscode/extensions.json](https://github.com/lobehub/lobe-chat/blob/main/.vscode/extensions.json) 中推荐安装的扩展获得最佳开发体验。
 
 ### 项目设置
 
@@ -32,14 +36,7 @@
 git clone https://github.com/lobehub/lobe-chat.git
 ```
 
-2. **安装依赖**：然后，进入项目目录，并使用 bun 安装项目的依赖包：
-
-```bash
-cd lobe-chat
-bun i
-```
-
-如果你使用 pnpm ，可以执行:
+2. **安装依赖**：然后，进入项目目录，并使用 `pnpm` 安装项目的依赖包：
 
 ```bash
 cd lobe-chat
@@ -54,7 +51,7 @@ bun run dev
 
 现在，你可以在浏览器中打开 `http://localhost:3010`，你应该能看到 LobeChat 的欢迎页面。这表明你已经成功地设置了开发环境。
 
-![](https://github-production-user-asset-6210df.s3.amazonaws.com/28616219/274655364-414bc31e-8511-47a3-af17-209b530effc7.png)
+![Chat Page](https://hub-apac-1.lobeobjects.space/docs/fc7b157a3bc016bc97719065f80c555c.png)
 
 在开发过程中，如果你在环境设置上遇到任何问题，或者有任何关于 LobeChat 开发的问题，欢迎随时向我们提问。我们期待看到你的贡献！
 

docs/development/basic/work-with-server-side-database.mdx

@@ -0,0 +1,77 @@
+# Work with Server-Side Database
+
+LobeChat provides a battery-included experience with its client-side database.
+While some features you really care about is only available at a server-side development.
+
+In order to work with the aspect of server-side database,
+you can setup all the prerequisites by following the [Deploying Server-Side Database](https://lobehub.com/docs/self-hosting/server-database) story.
+But here is the easier approach that can reduce your pain.
+
+## Quick Setup
+
+### Environment Configuration
+
+The project already includes a `.env.development` file with all necessary environment variables for server-side database mode. This file configures:
+
+- **Service Mode**: `NEXT_PUBLIC_SERVICE_MODE=server`
+- **Database**: PostgreSQL with connection string
+- **Authentication**: NextAuth with Casdoor SSO
+- **Storage**: MinIO S3-compatible storage
+- **Search**: SearXNG search engine
+
+### Start Docker Services
+
+Start all required services using Docker Compose:
+
+```bash
+docker-compose -f docker-compose.development.yml up -d
+```
+
+This will start the following services:
+
+- PostgreSQL database (port 5432)
+- MinIO storage (port 9000)
+- Casdoor authentication (port 8000)
+- SearXNG search (port 8080)
+
+### Run Database Migrations
+
+Execute the database migration script to create all necessary tables:
+
+```bash
+pnpm db:migrate
+```
+
+You should see: `✅ database migration pass.`
+
+### Start Development Server
+
+Launch the LobeChat development server:
+
+```bash
+pnpm dev
+```
+
+The server will start on `http://localhost:3010`
+
+And you can check all Docker services are running by running:
+
+```bash
+docker-compose -f docker-compose.development.yml ps
+```
+
+### Reset Services
+
+If you encounter issues, you can reset the entire stack:
+
+```bash
+# Stop and remove all containers
+docker-compose -f docker-compose.development.yml down
+
+# Remove volumes (this will delete all data)
+docker-compose -f docker-compose.development.yml down -v
+
+# Start fresh
+docker-compose -f docker-compose.development.yml up -d
+pnpm db:migrate
+```

docs/development/basic/work-with-server-side-database.zh-CN.mdx

@@ -0,0 +1,77 @@
+# 使用服务端数据库
+
+LobeChat 提供了内置的客户端数据库体验。
+但某些重要功能仅在服务端开发中可用。
+
+为了使用服务端数据库功能，
+需要参考 [部署服务端数据库](https://lobehub.com/docs/self-hosting/server-database) 的说明来配置所有前置条件。
+本文档提供了一个更简化的配置方法，能够在本地开发时快速启动简化的服务端环境。
+
+## 快速设置
+
+### 环境配置
+
+项目已经包含了一个 `.env.development` 文件，其中包含服务端数据库模式所需的所有环境变量。此文件配置了：
+
+- **服务模式**: `NEXT_PUBLIC_SERVICE_MODE=server`
+- **数据库**: 带连接字符串的 PostgreSQL
+- **身份验证**: 带 Casdoor SSO 的 NextAuth
+- **存储**: MinIO S3 兼容存储
+- **搜索**: SearXNG 搜索引擎
+
+### 启动 Docker 服务
+
+使用 Docker Compose 启动所有必需的服务：
+
+```bash
+docker-compose -f docker-compose.development.yml up -d
+```
+
+这将启动以下服务：
+
+- PostgreSQL 数据库（端口 5432）
+- MinIO 存储（端口 9000）
+- Casdoor 身份验证（端口 8000）
+- SearXNG 搜索（端口 8080）
+
+### 运行数据库迁移
+
+执行数据库迁移脚本以创建所有必要的表：
+
+```bash
+pnpm db:migrate
+```
+
+预期输出：`✅ database migration pass.`
+
+### 启动开发服务器
+
+启动 LobeChat 开发服务器：
+
+```bash
+pnpm dev
+```
+
+服务器将在 `http://localhost:3010` 上启动
+
+可以通过运行以下命令检查所有 Docker 服务运行状态：
+
+```bash
+docker-compose -f docker-compose.development.yml ps
+```
+
+### 重置服务
+
+如遇到问题，可以重置整个服务堆栈：
+
+```bash
+# 停止并删除所有容器
+docker-compose -f docker-compose.development.yml down
+
+# 删除卷（这将删除所有数据）
+docker-compose -f docker-compose.development.yml down -v
+
+# 重新启动
+docker-compose -f docker-compose.development.yml up -d
+pnpm db:migrate
+```

docs/development/database-schema.dbml

@@ -104,6 +104,19 @@ table ai_providers {
   }
 }
 
+table api_keys {
+  id integer [pk, not null]
+  name varchar(256) [not null]
+  key varchar(256) [not null, unique]
+  enabled boolean [default: true]
+  expires_at "timestamp with time zone"
+  last_used_at "timestamp with time zone"
+  user_id text [not null]
+  accessed_at "timestamp with time zone" [not null, default: `now()`]
+  created_at "timestamp with time zone" [not null, default: `now()`]
+  updated_at "timestamp with time zone" [not null, default: `now()`]
+}
+
 table async_tasks {
   id uuid [pk, not null, default: `gen_random_uuid()`]
   type text
@@ -116,6 +129,39 @@ table async_tasks {
   updated_at "timestamp with time zone" [not null, default: `now()`]
 }
 
+table chat_groups {
+  id text [pk, not null]
+  title text
+  description text
+  config jsonb
+  client_id text
+  user_id text [not null]
+  pinned boolean [default: false]
+  accessed_at "timestamp with time zone" [not null, default: `now()`]
+  created_at "timestamp with time zone" [not null, default: `now()`]
+  updated_at "timestamp with time zone" [not null, default: `now()`]
+
+  indexes {
+    (client_id, user_id) [name: 'chat_groups_client_id_user_id_unique', unique]
+  }
+}
+
+table chat_groups_agents {
+  chat_group_id text [not null]
+  agent_id text [not null]
+  user_id text [not null]
+  enabled boolean [default: true]
+  order integer [default: 0]
+  role text [default: 'participant']
+  accessed_at "timestamp with time zone" [not null, default: `now()`]
+  created_at "timestamp with time zone" [not null, default: `now()`]
+  updated_at "timestamp with time zone" [not null, default: `now()`]
+
+  indexes {
+    (chat_group_id, agent_id) [pk]
+  }
+}
+
 table document_chunks {
   document_id varchar(30) [not null]
   chunk_id uuid [not null]
@@ -359,6 +405,8 @@ table messages {
   parent_id text
   quota_id text
   agent_id text
+  group_id text
+  target_id text
   accessed_at "timestamp with time zone" [not null, default: `now()`]
   created_at "timestamp with time zone" [not null, default: `now()`]
   updated_at "timestamp with time zone" [not null, default: `now()`]
@@ -702,6 +750,7 @@ table rbac_roles {
   description text
   is_system boolean [not null, default: false]
   is_active boolean [not null, default: true]
+  metadata jsonb [default: `{}`]
   accessed_at "timestamp with time zone" [not null, default: `now()`]
   created_at "timestamp with time zone" [not null, default: `now()`]
   updated_at "timestamp with time zone" [not null, default: `now()`]
@@ -824,6 +873,7 @@ table topics {
   title text
   favorite boolean [default: false]
   session_id text
+  group_id text
   user_id text [not null]
   client_id text
   history_summary text
@@ -895,6 +945,14 @@ ref: agents_to_sessions.session_id > sessions.id
 
 ref: agents_to_sessions.agent_id > agents.id
 
+ref: chat_groups_agents.chat_group_id > chat_groups.id
+
+ref: chat_groups_agents.agent_id > agents.id
+
+ref: chat_groups_agents.user_id - users.id
+
+ref: chat_groups.user_id - users.id
+
 ref: unstructured_chunks.file_id - files.id
 
 ref: document_chunks.document_id > documents.id

docs/self-hosting/advanced/s3/cloudflare-r2.mdx

@@ -85,7 +85,7 @@ We need to configure an S3 storage service in the server-side database to store
   <Image alt={'Configure allowed site domain'} src={'https://github.com/lobehub/lobe-chat/assets/28616219/dfcc2cb3-2958-4498-a8a4-51bec584fe7d'} />
 
   <Callout type={'info'}>
-    If you also plan to use the desktop client, add <code>http://localhost:3015</code> to <code>AllowedOrigins</code> so the desktop client (running locally) can access R2.
+    If you also plan to use the desktop client, add <code>[http://localhost:3015](http://localhost:3015)</code> to <code>AllowedOrigins</code> so the desktop client (running locally) can access R2.
   </Callout>
 
   Example configuration is as follows:

docs/self-hosting/advanced/s3/cloudflare-r2.zh-CN.mdx

@@ -82,10 +82,10 @@ tags:
   添加跨域规则，允许你的域名（在上文是 `https://your-project.vercel.app`）来源的请求：
 
   <Image alt={'配置允许你的站点域名'} src={'https://github.com/lobehub/lobe-chat/assets/28616219/dfcc2cb3-2958-4498-a8a4-51bec584fe7d'} />
-  <Callout type={'info'}>
-    如果你还需要在桌面端使用，请在 <code>AllowedOrigins</code> 中额外添加 <code>http://localhost:3015</code>，以便桌面端（本地运行）能够访问 R2。
-  </Callout>
 
+  <Callout type={'info'}>
+    如果你还需要在桌面端使用，请在 <code>AllowedOrigins</code> 中额外添加 <code>[http://localhost:3015](http://localhost:3015)</code>，以便桌面端（本地运行）能够访问 R2。
+  </Callout>
 
   示例配置如下：
 

docs/self-hosting/advanced/webrtc.mdx

@@ -1,86 +0,0 @@
----
-title: LobeChat WebRTC Sync - Real-Time Data Sharing
-description: >-
-  Explore LobeChat's WebRTC sync for real-time data sharing and privacy without servers.
-
-tags:
-  - WebRTC
-  - LobeChat
-  - Data Synchronization
-  - Real-Time Communication
-  - Peer-to-Peer
----
-
-# LobeChat WebRTC Sync
-
-## Introduction to WebRTC
-
-WebRTC (Web Real-Time Communication) is a technology that enables peer-to-peer communication between browsers. In LobeChat, we experimentally implemented real-time data synchronization between devices based on WebRTC and YJS, without relying on traditional server databases. This solution offers high privacy, zero conflicts, and provides a real-time session synchronization experience.
-
-## Configuring WebRTC for Synchronization
-
-To use the WebRTC synchronization feature in LobeChat, you need to complete the following steps:
-
-<Steps>
-  ### Deploy Signaling Server
-
-  Deploy a WebRTC signaling server with one click using the Zeabur platform:
-
-  [![Deploy on Zeabur](https://zeabur.com/button.svg)](https://zeabur.com/templates/MY0JZG?referralCode=arvinxx)
-
-  Alternatively, you can view the [source code](https://github.com/lobehub/y-webrtc-signaling) and deploy it on your own.
-
-  After deployment, you will receive a URL, for example: `https://my-signaling-server.zeabur.app`.
-
-  ### Enable WebRTC Sync in the Deployment Instance
-
-  The WebRTC sync feature in LobeChat is hidden by default and needs to be enabled by adding the environment variable `FEATURE_FLAGS=+webrtc_sync`.
-
-  ### Configure WebRTC Sync Settings in LobeChat
-
-  1. Open LobeChat settings -> Data Sync
-  2. Enter the signaling server address in the WebRTC sync section;
-  3. Set the sync channel name and password
-
-  <Image alt={'LobeChat Data Sync Settings Page'} height={356} inStep src={'https://github.com/lobehub/lobe-chat/assets/28616219/bf86bf1e-87fb-4015-8587-15ff28bb9c24'} />
-
-  ### Repeat the Above Configuration on Devices that Need to Sync
-
-  Ensure all devices use the same signaling server, channel name, and password. Once configured, the devices should automatically start syncing data.
-</Steps>
-
-## Limitations and Known Issues
-
-Although WebRTC has the advantages of no database and flexibility, after extensive community testing, the following limitations and known issues have been identified:
-
-### Requirement for Devices to be Online Simultaneously
-
-WebRTC requires devices to be online simultaneously to synchronize, meaning changes cannot be made on one device while offline and then synced later on another device.
-
-This limitation is due to the communication nature of WebRTC. In a pure frontend, serverless scenario, data synchronization between two devices can only be achieved through peer-to-peer communication. When one device is online and the other is offline, it is impossible to determine where the data should come from. Only when both devices are online can data communication occur. This mode is more like an online chat room where everyone needs to be online to see each other's data and achieve synchronization.
-
-Therefore, in certain situations, WebRTC's pure peer-to-peer approach may not fully meet users' needs (e.g., one device is a work computer, and the other is a home computer), and there are also some issues with data synchronization.
-
-### Network Issues Leading to Sync Failures
-
-Due to the implementation mechanism of WebRTC, its peer-to-peer communication has strict network requirements. Many of our users have reported:
-
-- Syncing between PCs is possible, but syncing between a mobile device with a SIM card and a PC is not, although syncing is possible when using the same WIFI as the PC;
-- Syncing fails when switching networks.
-
-### Stability and Performance Issues
-
-- Some users have reported ICE connection failures on the Firefox browser: [WebRTC Data Sync Feedback](https://github.com/lobehub/lobe-chat/issues/1683#issuecomment-2094745907)
-- For extremely long text or large amounts of conversation records, the synchronization process may slow down or become unstable: [When the model outputs a very long conversation, the end of the conversation will contain synchronization-related content tags, leading to sync failures](https://github.com/lobehub/lobe-chat/issues/1962)
-
-## Our Recommendations
-
-Considering the above reasons, we recommend users treat the WebRTC sync feature as experimental and regularly back up important data.
-
-We have already released a more stable and user-friendly server database synchronization solution ([deployment guide](/docs/self-hosting/advanced/server-database)). We recommend users prioritize using the server database synchronization solution.
-
-<Callout type={'warning'}>
-  Please note that we have officially announced the archiving of this sync feature in [PR
-  3182](https://github.com/lobehub/lobe-chat/pull/3182), and the above issues will no longer be
-  considered for fixes.
-</Callout>

docs/self-hosting/advanced/webrtc.zh-CN.mdx

@@ -1,80 +0,0 @@
----
-title: LobeChat WebRTC 同步配置指南
-description: 在 LobeChat 中实现基于 WebRTC 和 YJS 的设备间实时数据同步。了解如何配置 WebRTC 并开启同步功能，以及使用局限性和已知问题。
-tags:
-  - YJS
-  - 信令服务器
----
-
-# LobeChat WebRTC 同步
-
-## WebRTC 简介
-
-WebRTC (Web Real-Time Communication) 是一项实现浏览器之间点对点通信的技术。在 LobeChat 中，我们实验性地基于 WebRTC 和 YJS 实现了设备间的实时数据同步，无需依赖传统的服务器数据库。这种方案具有高度隐私性、零冲突性，并能提供实时会话同步体验。
-
-## 配置 WebRTC 并实现同步
-
-要使用 LobeChat 的 WebRTC 同步功能，需要完成以下步骤:
-
-<Steps>
-  ### 部署信令服务器
-
-  使用 Zeabur 平台一键部署 WebRTC 信令服务器：
-
-  [![Deploy on Zeabur](https://zeabur.com/button.svg)](https://zeabur.com/templates/MY0JZG?referralCode=arvinxx)
-
-  或者查看 [源码](https://github.com/lobehub/y-webrtc-signaling) 自行部署。
-
-  部署完成后，可以得到一个 URL，例如：`https://my-signaling-server.zeabur.app`。
-
-  ### 在部署实例中开启 WebRTC 同步
-
-  LobeChat 默认隐藏了 WebRTC 同步功能，需要通过添加环境变量 `FEATURE_FLAGS=+webrtc_sync` 来开启 WebRTC 同步特性。
-
-  ### 配置 LobeChat 的 WebRTC 同步设置
-
-  1. 打开 LobeChat 设置 -> 数据同步
-  2. 在 WebRTC 同步中填写信令服务器地址；
-  3. 设置同步频道名称和密码
-
-  <Image alt={'LobeChat 数据同步设置页'} height={356} inStep src={'https://github.com/lobehub/lobe-chat/assets/28616219/bf86bf1e-87fb-4015-8587-15ff28bb9c24'} />
-
-  ### 在需要同步的设备上重复以上配置
-
-  确保所有设备使用相同的信令服务器、频道名称和密码，完成配置后，设备间应该可以开始自动同步数据。
-</Steps>
-
-## 使用局限性和已知问题
-
-虽然 WebRTC 具有无数据库、比较灵活的特性，但目前该功能经过大范围社区测试，存在以下局限性和已知问题：
-
-### 设备同时在线要求
-
-WebRTC 要求设备同时在线才能进行同步，这意味着无法在一台设备离线时在另一台设备上进行更改并稍后同步。
-
-这是 WebRTC 本身的通信特性有关系，由于在纯前端、无服务端的情况下，两个设备的数据同步只能通过点对点通信的形式达成。当一个设备在线，一个设备离线的情况下，我们无从感知数据到底应该从哪来，只有当两台设备都在线的时候，双发数据才能通信。其实这种模式更像是一个在线聊天室，大家都在线时才能看到对方的数据，然后达成同步。
-
-因此 WebRTC 这种纯点对点的方式在某些情况下并无法完全满足用户的诉求（例如一个是公司电脑，一个是家里电脑），同时也存在一些数据同步层面的问题。
-
-### 网络问题可能导致同步失败
-
-由于 WebRTC 的实现机制，其点对点通信对于网络要求非常苛刻，我们的很多用户反馈：
-
-- 在 PC 上可以互相同步、 手机 sim 卡无法和 PC 同步、但是换成和 PC 一样的 WIFI 可以和 PC 同步；
-- 任何切换网络都无法同步；
-
-### 稳定性与性能问题
-
-- 部分用户报告在 Firefox 浏览器上遇到 ICE 连接失败的问题：[WebRTC Data Sync Feedback](https://github.com/lobehub/lobe-chat/issues/1683#issuecomment-2094745907)
-- 对于超长文本或大量对话记录，同步过程可能变慢或不稳定：[当模型输出超长对话时，对话末尾会出现同步相关的内容标签，导致同步失败](https://github.com/lobehub/lobe-chat/issues/1962)
-
-## 我们的建议
-
-鉴于以上原因，我们建议用户将 WebRTC 同步功能视为实验性功能，并定期备份重要数据。
-
-目前我们已经发布了更稳定、更用户友好的服务端数据库同步方案（[部署指南](/zh/docs/self-hosting/advanced/server-database)），我们建议用户优先考虑使用服务端数据库同步方案。
-
-<Callout type={'warning'}>
-  请注意，我们已经在 [PR 3182](https://github.com/lobehub/lobe-chat/pull/3182)
-  中正式宣布归档该同步特性，上述问题将不再考虑进行修复。
-</Callout>

docs/self-hosting/environment-variables/basic.mdx

@@ -37,13 +37,6 @@ When using the `random` mode, a random API Key will be selected from the availab
 
 When using the `turn` mode, the API Keys will be retrieved in a polling manner according to the specified order.
 
-### `NEXT_PUBLIC_BASE_PATH`
-
-- Type: Optional
-- Description: Add a `basePath` for LobeChat.
-- Default: -
-- Example: `/test`
-
 ### `DEFAULT_AGENT_CONFIG`
 
 - Type: Optional

docs/self-hosting/environment-variables/basic.zh-CN.mdx

@@ -34,13 +34,6 @@ LobeChat 在部署时提供了一些额外的配置项，你可以使用环境
 
 使用 `turn` 模式下，将按照填写的顺序，轮询获取得到 API Key。
 
-### `NEXT_PUBLIC_BASE_PATH`
-
-- 类型：可选
-- 描述：为 LobeChat 添加 `basePath`
-- 默认值: `-`
-- 示例: `/test`
-
 ### `DEFAULT_AGENT_CONFIG`
 
 - 类型：可选

docs/self-hosting/environment-variables/model-provider.mdx

@@ -646,8 +646,33 @@ If you need to use Azure OpenAI to provide model services, you can refer to the
 - Type: Optional
 - Description: Used to control the FAL model list. Use `+` to add a model, `-` to hide a model, and `model_name=display_name` to customize the display name of a model. Separate multiple entries with commas. The definition syntax follows the same rules as other providers' model lists.
 - Default: `-`
-- Example: `-all,+fal-model-1,+fal-model-2=fal-special`
+- Example: `-all,+flux/schnell,+flux-pro/kontext=FLUX.1 Kontext [pro]`
 
-The above example disables all models first, then enables `fal-model-1` and `fal-model-2` (displayed as `fal-special`).
+The above example disables all models first, then enables `flux/schnell` and `flux-pro/kontext` (displayed as `FLUX.1 Kontext [pro]`).
+
+## BFL
+
+### `ENABLED_BFL`
+
+- Type: Optional
+- Description: Enables BFL as a model provider by default. Set to `0` to disable the BFL service.
+- Default: `1`
+- Example: `0`
+
+### `BFL_API_KEY`
+
+- Type: Required
+- Description: This is the API key you applied for in the BFL service.
+- Default: -
+- Example: `12345678-1234-1234-1234-123456789abc`
+
+### `BFL_MODEL_LIST`
+
+- Type: Optional
+- Description: Used to control the BFL model list. Use `+` to add a model, `-` to hide a model, and `model_name=display_name` to customize the display name of a model. Separate multiple entries with commas. The definition syntax follows the same rules as other providers' model lists.
+- Default: `-`
+- Example: `-all,+flux-pro-1.1,+flux-kontext-pro=FLUX.1 Kontext [pro]`
+
+The above example disables all models first, then enables `flux-pro-1.1` and `flux-kontext-pro` (displayed as `FLUX.1 Kontext [pro]`).
 
 [model-list]: /docs/self-hosting/advanced/model-list

docs/self-hosting/environment-variables/model-provider.zh-CN.mdx

@@ -645,8 +645,33 @@ LobeChat 在部署时提供了丰富的模型服务商相关的环境变量，
 - 类型：可选
 - 描述：用来控制 FAL 模型列表，使用 `+` 增加一个模型，使用 `-` 来隐藏一个模型，使用 `模型名=展示名` 来自定义模型的展示名，用英文逗号隔开。模型定义语法规则与其他 provider 保持一致。
 - 默认值：`-`
-- 示例：`-all,+fal-model-1,+fal-model-2=fal-special`
+- 示例：`-all,+flux/schnell,+flux-pro/kontext=FLUX.1 Kontext [pro]`
 
-上述示例表示先禁用所有模型，再启用 `fal-model-1` 和 `fal-model-2`（显示名为 `fal-special`）。
+上述示例表示先禁用所有模型，再启用 `flux/schnell` 和 `flux-pro/kontext`（显示名为 `FLUX.1 Kontext [pro]`）。
+
+## BFL
+
+### `ENABLED_BFL`
+
+- 类型：可选
+- 描述：默认启用 BFL 作为模型供应商，当设为 0 时关闭 BFL 服务
+- 默认值：`1`
+- 示例：`0`
+
+### `BFL_API_KEY`
+
+- 类型：必选
+- 描述：这是你在 BFL 服务中申请的 API 密钥
+- 默认值：-
+- 示例：`12345678-1234-1234-1234-123456789abc`
+
+### `BFL_MODEL_LIST`
+
+- 类型：可选
+- 描述：用来控制 BFL 模型列表，使用 `+` 增加一个模型，使用 `-` 来隐藏一个模型，使用 `模型名=展示名` 来自定义模型的展示名，用英文逗号隔开。模型定义语法规则与其他 provider 保持一致。
+- 默认值：`-`
+- 示例：`-all,+flux-pro-1.1,+flux-kontext-pro=FLUX.1 Kontext [pro]`
+
+上述示例表示先禁用所有模型，再启用 `flux-pro-1.1` 和 `flux-kontext-pro`（显示名为 `FLUX.1 Kontext [pro]`）。
 
 [model-list]: /zh/docs/self-hosting/advanced/model-list

docs/usage/providers/bfl.mdx

@@ -0,0 +1,68 @@
+---
+title: Using Black Forest Labs API Key in LobeChat
+description: >-
+  Learn how to integrate Black Forest Labs API Key in LobeChat for AI image generation using advanced models and high-quality output.
+
+tags:
+  - Black Forest Labs
+  - Image Generation
+  - API Key
+  - Web UI
+---
+
+# Using Black Forest Labs in LobeChat
+
+<Image alt={'Using Black Forest Labs in LobeChat'} cover src={'https://hub-apac-1.lobeobjects.space/docs/c7274a5b037227f042ad7558f535b5ea.png'} />
+
+[Black Forest Labs](https://bfl.ai/) is currently the world's top-tier AI image generation research lab, having developed the FLUX series of high-quality image generation models and the FLUX Kontext series of image editing models. This document will guide you on how to use Black Forest Labs in LobeChat:
+
+<Steps>
+  ### Step 1: Obtain Black Forest Labs API Key
+
+  - Register for a [Black Forest Labs account](https://auth.bfl.ai/).
+  - Navigate to the [API Keys dashboard](https://dashboard.bfl.ai/api/keys) and click **Add Key** to generate a new API key.
+  - Copy the generated API key and keep it secure; it will only be shown once.
+
+  <Image
+    alt={'Open the API dashboard'}
+    inStep
+    src={
+'https://hub-apac-1.lobeobjects.space/docs/36331f581e36a72ba91fb6c1dacbedd4.png'
+}
+  />
+
+  <Image
+    alt={'Create API Key'}
+    inStep
+    src={
+'https://hub-apac-1.lobeobjects.space/docs/3049ad79746bf913d1b2736cfe1d38fa.png'
+}
+  />
+
+  <Image
+    alt={'Retrieve API Key'}
+    inStep
+    src={
+'https://hub-apac-1.lobeobjects.space/docs/33143deb04f2a0cdb2fd543ff8161633.png'
+}
+  />
+
+  ### Step 2: Configure Black Forest Labs in LobeChat
+
+  - Visit the `Settings` page in LobeChat.
+  - Under **AI Service Provider**, locate the **Black Forest Labs** configuration section.
+
+  <Image alt={'Enter API Key'} inStep src={'https://hub-apac-1.lobeobjects.space/docs/4fe1993d94fad72829612d22527a2ecd.png'} />
+
+  - Paste the API key you obtained.
+  - Choose a Black Forest Labs model for image generation.
+
+  <Image alt={'Select Black Forest Labs model for image generation'} inStep src={'https://hub-apac-1.lobeobjects.space/docs/7a6dfdb1a1bdd5443104a6d2cb3a8bf4.png'} />
+
+  <Callout type={'warning'}>
+    During usage, you may incur charges according to Black Forest Labs's pricing policy. Please review Black Forest Labs's
+    official pricing before heavy usage.
+  </Callout>
+</Steps>
+
+You can now use Black Forest Labs's advanced image generation models directly within LobeChat to create stunning visual content.

docs/usage/providers/bfl.zh-CN.mdx

@@ -0,0 +1,67 @@
+---
+title: 在 LobeChat 中使用 Black Forest Labs API Key
+description: >-
+  学习如何在 LobeChat 中配置和使用 Black Forest Labs API Key，使用先进模型进行高质量 AI 图像生成。
+
+tags:
+  - Black Forest Labs
+  - 图像生成
+  - API Key
+  - Web UI
+---
+
+# 在 LobeChat 中使用 Black Forest Labs
+
+<Image alt={'在 LobeChat 中使用 Black Forest Labs'} cover src={'https://hub-apac-1.lobeobjects.space/docs/c7274a5b037227f042ad7558f535b5ea.png'} />
+
+[Black Forest Labs](https://bfl.ai/) 是当前世界最顶级的 AI 图像生成实验室团队，研发了 FLUX 系列高质量图像生成模型，FLUX Kontext 系列图像编辑模型。本文将指导你如何在 LobeChat 中使用 Black Forest Labs：
+
+<Steps>
+  ### 步骤一：获取 Black Forest Labs API Key
+
+  - 注册 [Black Forest Labs](https://auth.bfl.ai/) 账户；
+  - 前往 [API Keys 控制台](https://dashboard.bfl.ai/api/keys)，点击 **Add Key** 创建新的 API 密钥；
+  - 复制生成的 API Key 并妥善保存，它只会显示一次。
+
+  <Image
+    alt={'打开 API 控制台'}
+    inStep
+    src={
+'https://hub-apac-1.lobeobjects.space/docs/36331f581e36a72ba91fb6c1dacbedd4.png'
+}
+  />
+
+  <Image
+    alt={'创建 API Key'}
+    inStep
+    src={
+'https://hub-apac-1.lobeobjects.space/docs/3049ad79746bf913d1b2736cfe1d38fa.png'
+}
+  />
+
+  <Image
+    alt={'获取 API Key'}
+    inStep
+    src={
+'https://hub-apac-1.lobeobjects.space/docs/33143deb04f2a0cdb2fd543ff8161633.png'
+}
+  />
+
+  ### 步骤二：在 LobeChat 中配置 Black Forest Labs
+
+  - 访问 LobeChat 的 `设置` 页面；
+  - 在 `AI服务商` 下找到 `Black Forest Labs` 的设置项；
+
+  <Image alt={'填入 API 密钥'} inStep src={'https://hub-apac-1.lobeobjects.space/docs/4fe1993d94fad72829612d22527a2ecd.png'} />
+
+  - 粘贴获取到的 API Key；
+  - 选择一个 Black Forest Labs 模型用于图像生成。
+
+  <Image alt={'选择 Black Forest Labs 模型进行图像生成'} inStep src={'https://hub-apac-1.lobeobjects.space/docs/7a6dfdb1a1bdd5443104a6d2cb3a8bf4.png'} />
+
+  <Callout type={'warning'}>
+    在使用过程中，你可能需要向 Black Forest Labs 支付相应费用，请在大量调用前查阅 Black Forest Labs 的官方计费政策。
+  </Callout>
+</Steps>
+
+至此，你已经可以在 LobeChat 中使用 Black Forest Labs 提供的先进图像生成模型来创作精美的视觉内容了。

drizzle.config.ts

@@ -22,8 +22,8 @@ export default {
     url: connectionString,
   },
   dialect: 'postgresql',
-  out: './src/database/migrations',
+  out: './packages/database/migrations',
 
-  schema: './src/database/schemas',
+  schema: './packages/database/src/schemas',
   strict: true,
 } satisfies Config;

locales/ar/chat.json

@@ -52,7 +52,11 @@
     "thinking": {
       "title": "مفتاح التفكير العميق"
     },
-    "title": "وظائف توسيع النموذج"
+    "title": "وظائف توسيع النموذج",
+    "urlContext": {
+      "desc": "عند التفعيل، سيتم تحليل روابط الويب تلقائيًا للحصول على محتوى السياق الفعلي للصفحة",
+      "title": "استخراج محتوى رابط الويب"
+    }
   },
   "history": {
     "title": "سيتذكر المساعد آخر {{count}} رسالة فقط"
@@ -125,6 +129,7 @@
       "inputWriteCached": "تخزين إدخال الكتابة",
       "output": "مخرجات",
       "outputAudio": "مخرجات صوتية",
+      "outputImage": "إخراج الصورة",
       "outputText": "مخرجات نصية",
       "outputTitle": "تفاصيل المخرجات",
       "reasoning": "تفكير عميق",
@@ -268,7 +273,8 @@
     "clientMode": {
       "actionFiletip": "رفع ملف",
       "actionTooltip": "رفع",
-      "disabled": "النموذج الحالي لا يدعم التعرف على الصور وتحليل الملفات، يرجى تغيير النموذج لاستخدامه"
+      "disabled": "النموذج الحالي لا يدعم التعرف على الصور وتحليل الملفات، يرجى تغيير النموذج لاستخدامه",
+      "visionNotSupported": "النموذج الحالي لا يدعم التعرف البصري، يرجى تبديل النموذج لاستخدام هذه الميزة"
     },
     "preview": {
       "prepareTasks": "تحضير الأجزاء...",

locales/ar/common.json

@@ -32,6 +32,7 @@
   "blog": "مدونة المنتجات",
   "branching": "إنشاء موضوع فرعي",
   "branchingDisable": "ميزة \"الموضوع الفرعي\" متاحة فقط في إصدار الخادم. إذا كنت بحاجة إلى هذه الميزة، يرجى التبديل إلى وضع نشر الخادم أو استخدام LobeChat Cloud",
+  "branchingRequiresSavedTopic": "الموضوع الحالي غير محفوظ، يجب الحفظ قبل استخدام ميزة الموضوع الفرعي",
   "cancel": "إلغاء",
   "changelog": "سجل التغييرات",
   "clientDB": {
@@ -182,6 +183,13 @@
     "title": "هل تحب منتجنا؟"
   },
   "fullscreen": "وضع كامل الشاشة",
+  "geminiImageChineseWarning": {
+    "content": "قد يفشل Nano Banana أحيانًا في إنشاء الصور عند استخدام اللغة الصينية. يُنصح باستخدام اللغة الإنجليزية للحصول على نتائج أفضل.",
+    "continueGenerate": "متابعة الإنشاء",
+    "continueSend": "متابعة الإرسال",
+    "doNotShowAgain": "عدم الإظهار مرة أخرى",
+    "title": "تنبيه إدخال اللغة الصينية"
+  },
   "historyRange": "نطاق التاريخ",
   "import": "استيراد",
   "importData": "استيراد البيانات",

locales/ar/components.json

@@ -1,4 +1,9 @@
 {
+  "ArgsInput": {
+    "addArgument": "إضافة معامل",
+    "argumentPlaceholder": "المعامل {{index}}",
+    "enterFirstArgument": "أدخل المعامل الأول..."
+  },
   "DragUpload": {
     "dragDesc": "اسحب الملفات هنا، يدعم تحميل عدة صور.",
     "dragFileDesc": "اسحب الصور والملفات هنا، يدعم تحميل عدة صور وملفات.",
@@ -68,6 +73,18 @@
   "GoBack": {
     "back": "عودة"
   },
+  "HtmlPreview": {
+    "actions": {
+      "download": "تنزيل",
+      "preview": "معاينة"
+    },
+    "iframeTitle": "معاينة HTML",
+    "mode": {
+      "code": "رمز",
+      "preview": "معاينة"
+    },
+    "title": "معاينة HTML"
+  },
   "ImageUpload": {
     "actions": {
       "changeImage": "انقر لتغيير الصورة",
@@ -125,6 +142,12 @@
     },
     "progress": {
       "uploadingWithCount": "تم تحميل {{completed}} من أصل {{total}}"
+    },
+    "validation": {
+      "fileSizeExceeded": "تجاوز حجم الملف الحد المسموح به",
+      "fileSizeExceededDetail": "{{fileName}} ({{actualSize}}) يتجاوز الحد الأقصى للحجم وهو {{maxSize}}",
+      "fileSizeExceededMultiple": "{{count}} من الملفات تتجاوز الحد الأقصى للحجم {{maxSize}}: {{fileList}}",
+      "imageCountExceeded": "تجاوز عدد الصور الحد المسموح به"
     }
   },
   "OllamaSetupGuide": {

locales/ar/error.json

@@ -85,6 +85,17 @@
     "CreateMessageError": "عذرًا، لم يتم إرسال الرسالة بشكل صحيح، يرجى نسخ المحتوى وإعادة إرساله، بعد تحديث الصفحة لن يتم الاحتفاظ بهذه الرسالة",
     "ExceededContextWindow": "المحتوى المطلوب الحالي يتجاوز الطول الذي يمكن للنموذج معالجته، يرجى تقليل كمية المحتوى ثم إعادة المحاولة",
     "FreePlanLimit": "أنت حاليًا مستخدم مجاني، لا يمكنك استخدام هذه الوظيفة، يرجى الترقية إلى خطة مدفوعة للمتابعة",
+    "GoogleAIBlockReason": {
+      "BLOCKLIST": "يحتوي المحتوى الذي أرسلته على كلمات محظورة. يرجى مراجعته وتعديل مدخلاتك ثم المحاولة مرة أخرى.",
+      "IMAGE_SAFETY": "تم حظر المحتوى الصوري الناتج لأسباب تتعلق بالأمان. يرجى محاولة تعديل طلب توليد الصورة.",
+      "LANGUAGE": "اللغة التي تستخدمها غير مدعومة مؤقتًا. يرجى المحاولة باللغة الإنجليزية أو بلغة أخرى مدعومة.",
+      "OTHER": "تم حظر المحتوى لسبب غير معروف. يرجى محاولة إعادة صياغة طلبك.",
+      "PROHIBITED_CONTENT": "قد يحتوي طلبك على محتوى محظور. يرجى تعديل طلبك لضمان توافقه مع سياسات الاستخدام.",
+      "RECITATION": "تم حظر محتواك لكونه قد ينتهك حقوق النشر. يرجى محاولة استخدام محتوى أصلي أو إعادة صياغة طلبك.",
+      "SAFETY": "تم حظر المحتوى بسبب سياسات السلامة. يرجى تعديل طلبك لتجنب أي محتوى ضار أو غير مناسب.",
+      "SPII": "قد يحتوي المحتوى على معلومات شخصية حساسة. لحماية الخصوصية، يرجى إزالة المعلومات الحساسة ثم المحاولة مرة أخرى.",
+      "default": "تم حظر المحتوى: {{blockReason}}. يرجى تعديل طلبك ثم المحاولة مرة أخرى."
+    },
     "InsufficientQuota": "عذرًا، لقد reached الحد الأقصى للحصة (quota) لهذه المفتاح، يرجى التحقق من رصيد الحساب الخاص بك أو زيادة حصة المفتاح ثم المحاولة مرة أخرى",
     "InvalidAccessCode": "كلمة المرور غير صحيحة أو فارغة، يرجى إدخال كلمة مرور الوصول الصحيحة أو إضافة مفتاح API مخصص",
     "InvalidBedrockCredentials": "فشلت مصادقة Bedrock، يرجى التحقق من AccessKeyId/SecretAccessKey وإعادة المحاولة",

locales/ar/modelProvider.json

@@ -153,6 +153,28 @@
     "searchProviders": "البحث عن مزودين...",
     "sort": "ترتيب مخصص"
   },
+  "newapi": {
+    "apiKey": {
+      "desc": "مفتاح API المقدم من منصة New API",
+      "placeholder": "مفتاح API الخاص بـ New API",
+      "required": "مفتاح API مطلوب",
+      "title": "مفتاح API"
+    },
+    "apiUrl": {
+      "desc": "عنوان API لخدمة New API، غالبًا ما يحتاج إلى /v1",
+      "title": "عنوان API"
+    },
+    "enabled": {
+      "title": "تمكين New API"
+    },
+    "models": {
+      "batchSelect": "تحديد نماذج متعددة ({{count}})",
+      "fetch": "جلب قائمة النماذج",
+      "selected": "النماذج المختارة",
+      "title": "النماذج المتاحة"
+    },
+    "title": "New API"
+  },
   "ollama": {
     "checker": {
       "desc": "اختبر ما إذا تم إدخال عنوان الوكيل بشكل صحيح",
@@ -185,6 +207,10 @@
     }
   },
   "providerModels": {
+    "batchSelect": {
+      "selected": "تم اختيار {{count}} نموذج",
+      "title": "تحديد متعدد"
+    },
     "config": {
       "aesGcm": "سيتم استخدام خوارزمية التشفير <1>AES-GCM</1> لتشفير مفتاحك وعنوان الوكيل وما إلى ذلك",
       "apiKey": {

locales/ar/models.json

@@ -74,9 +74,15 @@
   "DeepSeek-V3": {
     "description": "DeepSeek-V3 هو نموذج MoE تم تطويره ذاتيًا بواسطة شركة DeepSeek. حقق DeepSeek-V3 نتائج تقييم تفوقت على نماذج مفتوحة المصدر الأخرى مثل Qwen2.5-72B و Llama-3.1-405B، وفي الأداء ينافس النماذج المغلقة الرائدة عالميًا مثل GPT-4o و Claude-3.5-Sonnet."
   },
+  "DeepSeek-V3-1": {
+    "description": "DeepSeek V3.1: نموذج الاستدلال من الجيل التالي، يعزز القدرات على الاستدلال المعقد والتفكير التسلسلي، مناسب للمهام التي تتطلب تحليلاً عميقاً."
+  },
   "DeepSeek-V3-Fast": {
     "description": "مزود النموذج: منصة sophnet. DeepSeek V3 Fast هو النسخة السريعة عالية TPS من إصدار DeepSeek V3 0324، غير مكوّن بالكامل، يتمتع بقدرات برمجية ورياضية أقوى واستجابة أسرع!"
   },
+  "DeepSeek-V3.1": {
+    "description": "DeepSeek-V3.1 هو نموذج استدلال هجين جديد أطلقته DeepSeek، يدعم وضعين للاستدلال: التفكير وعدم التفكير، مع كفاءة تفكير أعلى مقارنة بـ DeepSeek-R1-0528. بعد تحسين ما بعد التدريب، تم تعزيز استخدام أدوات الوكيل وأداء مهام الوكيل بشكل كبير."
+  },
   "Doubao-lite-128k": {
     "description": "Doubao-lite يتميز بسرعة استجابة فائقة وقيمة أفضل مقابل المال، ويوفر خيارات أكثر مرونة للعملاء في سيناريوهات مختلفة. يدعم الاستدلال والتخصيص مع نافذة سياق 128k."
   },
@@ -131,9 +137,15 @@
   "ERNIE-Speed-Pro-128K": {
     "description": "نموذج اللغة الكبير عالي الأداء الذي طورته بايدو، والذي تم إصداره في عام 2024، يتمتع بقدرات عامة ممتازة، ويتميز بأداء أفضل من ERNIE Speed، مناسب كنموذج أساسي للتعديل الدقيق، مما يساعد على معالجة مشكلات السيناريوهات المحددة بشكل أفضل، مع أداء استدلال ممتاز."
   },
+  "FLUX-1.1-pro": {
+    "description": "FLUX.1.1 Pro"
+  },
   "FLUX.1-Kontext-dev": {
     "description": "FLUX.1-Kontext-dev هو نموذج متعدد الوسائط لتوليد وتحرير الصور طورته Black Forest Labs، يعتمد على بنية Rectified Flow Transformer ويحتوي على 12 مليار معلمة، يركز على توليد وإعادة بناء وتعزيز أو تحرير الصور بناءً على شروط سياقية محددة. يجمع النموذج بين مزايا التوليد القابل للتحكم في نماذج الانتشار وقدرات نمذجة السياق في Transformer، ويدعم إخراج صور عالية الجودة، ويستخدم على نطاق واسع في إصلاح الصور، إكمال الصور، وإعادة بناء المشاهد البصرية."
   },
+  "FLUX.1-Kontext-pro": {
+    "description": "FLUX.1 Kontext [pro]"
+  },
   "FLUX.1-dev": {
     "description": "FLUX.1-dev هو نموذج لغة متعدد الوسائط مفتوح المصدر طورته Black Forest Labs، مُحسّن لمهام النص والصورة، يدمج قدرات فهم وتوليد الصور والنصوص. يعتمد على نماذج اللغة الكبيرة المتقدمة مثل Mistral-7B، ويحقق معالجة متزامنة للنص والصورة واستدلالًا معقدًا من خلال مشفر بصري مصمم بعناية وضبط دقيق متعدد المراحل."
   },
@@ -170,6 +182,9 @@
   "Llama-3.2-90B-Vision-Instruct\t": {
     "description": "قدرات استدلال الصور المتقدمة المناسبة لتطبيقات الوكلاء في الفهم البصري."
   },
+  "Meta-Llama-3-3-70B-Instruct": {
+    "description": "Llama 3.3 70B: نموذج Transformer متعدد الاستخدامات، مناسب للحوار ومهام التوليد."
+  },
   "Meta-Llama-3.1-405B-Instruct": {
     "description": "نموذج نصي تم تعديله تحت الإشراف من Llama 3.1، تم تحسينه لحالات الحوار متعددة اللغات، حيث يتفوق في العديد من نماذج الدردشة مفتوحة ومغلقة المصدر المتاحة في المعايير الصناعية الشائعة."
   },
@@ -188,6 +203,9 @@
   "Meta-Llama-3.3-70B-Instruct": {
     "description": "Llama 3.3 هو النموذج اللغوي مفتوح المصدر متعدد اللغات الأكثر تقدمًا في سلسلة Llama، حيث يقدم تجربة تنافس أداء نموذج 405B بتكلفة منخفضة للغاية. يعتمد على هيكل Transformer، وتم تحسين فائدته وأمانه من خلال التعديل الدقيق تحت الإشراف (SFT) والتعلم المعزز من خلال ردود الفعل البشرية (RLHF). تم تحسين إصدار التعديل الخاص به ليكون مثاليًا للحوار متعدد اللغات، حيث يتفوق في العديد من المعايير الصناعية على العديد من نماذج الدردشة مفتوحة ومغلقة المصدر. تاريخ انتهاء المعرفة هو ديسمبر 2023."
   },
+  "Meta-Llama-4-Maverick-17B-128E-Instruct-FP8": {
+    "description": "Llama 4 Maverick: نموذج واسع النطاق قائم على مزيج من الخبراء، يوفر استراتيجية تفعيل خبراء فعالة لتحقيق أداء متميز في الاستدلال."
+  },
   "MiniMax-M1": {
     "description": "نموذج استدلال جديد مطور ذاتيًا. رائد عالميًا: 80 ألف سلسلة تفكير × 1 مليون إدخال، أداء يضاهي أفضل النماذج العالمية."
   },
@@ -257,6 +275,9 @@
   "Pro/deepseek-ai/DeepSeek-V3": {
     "description": "DeepSeek-V3 هو نموذج لغوي مختلط الخبراء (MoE) يحتوي على 6710 مليار معلمة، يستخدم الانتباه المتعدد الرؤوس (MLA) وهيكل DeepSeekMoE، ويجمع بين استراتيجيات توازن الحمل بدون خسائر مساعدة، مما يحسن كفاءة الاستدلال والتدريب. تم تدريبه مسبقًا على 14.8 تريليون توكن عالية الجودة، وتم إجراء تعديل دقيق تحت الإشراف والتعلم المعزز، مما يجعل DeepSeek-V3 يتفوق على نماذج مفتوحة المصدر الأخرى، ويقترب من النماذج المغلقة الرائدة."
   },
+  "Pro/deepseek-ai/DeepSeek-V3.1": {
+    "description": "DeepSeek-V3.1 هو نموذج لغة كبير بنمط هجين أصدرته DeepSeek AI، وقد شهد ترقيات مهمة متعددة مقارنة بالإصدارات السابقة. من الابتكارات الرئيسية في هذا النموذج دمج \"وضع التفكير\" و\"وضع عدم التفكير\" في نموذج واحد، حيث يمكن للمستخدمين التبديل بينهما بسهولة عبر تعديل قالب المحادثة لتلبية متطلبات المهام المختلفة. من خلال تحسينات ما بعد التدريب المخصصة، تم تعزيز أداء V3.1 في استدعاء الأدوات ومهام الوكيل بشكل ملحوظ، مما يمكنه من دعم أدوات البحث الخارجية وتنفيذ مهام معقدة متعددة الخطوات بشكل أفضل. يعتمد النموذج على DeepSeek-V3.1-Base مع تدريب إضافي، حيث تم توسيع حجم بيانات التدريب بشكل كبير عبر طريقة التوسيع النصي الطويل على مرحلتين، مما يحسن أدائه في معالجة المستندات الطويلة والرموز البرمجية الطويلة. كنموذج مفتوح المصدر، يظهر DeepSeek-V3.1 قدرة تنافسية مع أفضل النماذج المغلقة في مجالات الترميز والرياضيات والاستدلال، وبفضل هيكله المختلط للخبراء (MoE)، يحافظ على سعة نموذج ضخمة مع تقليل تكلفة الاستدلال بفعالية."
+  },
   "Pro/moonshotai/Kimi-K2-Instruct": {
     "description": "Kimi K2 هو نموذج أساسي يعتمد على بنية MoE مع قدرات قوية في البرمجة والوكيل، يحتوي على 1 تريليون معلمة و32 مليار معلمة مفعلة. يتفوق نموذج K2 في اختبارات الأداء الأساسية في مجالات المعرفة العامة، البرمجة، الرياضيات والوكيل مقارنة بالنماذج المفتوحة المصدر الأخرى."
   },
@@ -332,12 +353,21 @@
   "Qwen/Qwen3-30B-A3B-Instruct-2507": {
     "description": "Qwen3-30B-A3B-Instruct-2507 هو نسخة محدثة من Qwen3-30B-A3B في وضع عدم التفكير. هذا نموذج خبير مختلط (MoE) يحتوي على 30.5 مليار معلمة إجمالية و3.3 مليار معلمة تنشيط. تم تعزيز النموذج بشكل كبير في عدة جوانب، بما في ذلك تحسين كبير في الالتزام بالتعليمات، والتفكير المنطقي، وفهم النصوص، والرياضيات، والعلوم، والبرمجة، واستخدام الأدوات. كما حقق تقدمًا ملموسًا في تغطية المعرفة متعددة اللغات، ويستطيع التوافق بشكل أفضل مع تفضيلات المستخدم في المهام الذاتية والمفتوحة، مما يمكنه من توليد ردود أكثر فائدة ونصوص ذات جودة أعلى. بالإضافة إلى ذلك، تم تعزيز قدرة النموذج على فهم النصوص الطويلة إلى 256 ألف رمز. هذا النموذج يدعم فقط وضع عدم التفكير، ولن ينتج علامات `<think></think>` في مخرجاته."
   },
+  "Qwen/Qwen3-30B-A3B-Thinking-2507": {
+    "description": "Qwen3-30B-A3B-Thinking-2507 هو أحدث نموذج من سلسلة Qwen3 أصدره فريق Tongyi Qianwen في علي بابا. كنموذج خبير مختلط (MoE) يضم إجمالي 305 亿 (30.5 مليار) معلمة و33 亿 (3.3 مليار) معلمة تنشيط، يركز على تعزيز قدرة معالجة المهام المعقدة. أظهر هذا النموذج تحسناً ملحوظاً في الاستدلال المنطقي والرياضيات والعلوم والبرمجة والاختبارات المعيارية الأكاديمية التي تتطلب خبرة بشرية متخصصة. وفي الوقت نفسه، تحسنت قدراته العامة بشكل كبير في الالتزام بالتعليمات واستخدام الأدوات وتوليد النصوص ومحاذاة التفضيلات البشرية. يدعم النموذج أصلاً فهم سياق طويل بطول 256K وقابل للتوسع ليصل إلى مليون توكن. تم تصميم هذا الإصدار لوضع \"التفكير\"، ويهدف إلى حل المهام شديدة التعقيد من خلال استدلال تفصيلي خطوة بخطوة، كما برزت قدرات الوكلاء (Agent) فيه بشكل مميز."
+  },
   "Qwen/Qwen3-32B": {
     "description": "Qwen3 هو نموذج جديد من الجيل التالي مع تحسينات كبيرة في القدرات، حيث يصل إلى مستويات رائدة في الاستدلال، المهام العامة، الوكلاء، واللغات المتعددة، ويدعم تبديل وضع التفكير."
   },
   "Qwen/Qwen3-8B": {
     "description": "Qwen3 هو نموذج جديد من الجيل التالي مع تحسينات كبيرة في القدرات، حيث يصل إلى مستويات رائدة في الاستدلال، المهام العامة، الوكلاء، واللغات المتعددة، ويدعم تبديل وضع التفكير."
   },
+  "Qwen/Qwen3-Coder-30B-A3B-Instruct": {
+    "description": "Qwen3-Coder-30B-A3B-Instruct هو نموذج شيفرة من سلسلة Qwen3 طوّرته فريق Tongyi Qianwen التابع لشركة علي بابا. كنموذج مُبسّط ومُحسّن، يركز على تعزيز قدرات معالجة الشيفرة مع الحفاظ على أداء وكفاءة عاليتين. يُظهر هذا النموذج مزايا أداء بارزة بين النماذج مفتوحة المصدر في مهام معقدة مثل البرمجة الوكِيلية (Agentic Coding)، التشغيل الآلي لمتصفحات الويب واستدعاء الأدوات. يدعم النموذج بطبيعته سياقاً طويلاً يصل إلى 256K توكن، وقابلاً للتوسيع حتى 1M توكن، مما يتيح فهماً ومعالجة أفضل على مستوى مستودعات الشيفرة. بالإضافة إلى ذلك، يوفر النموذج دعماً قوياً للترميز الوكيل على منصات مثل Qwen Code وCLINE، وقد صُمّم بصيغة استدعاء دوال مخصّصة."
+  },
+  "Qwen/Qwen3-Coder-480B-A35B-Instruct": {
+    "description": "Qwen3-Coder-480B-A35B-Instruct هو نموذج برمجي أطلقته شركة علي بابا، ويعد حتى الآن الأكثر قدرةً على العمل كوكيل (Agentic). إنه نموذج مختلط الخبراء (Mixture-of-Experts, MoE) يضم 480 مليار معامل إجماليًا و35 مليار معامل نشط، محققًا توازنًا بين الكفاءة والأداء. يدعم النموذج بشكل أصلي طول سياق يصل إلى 256K (حوالي 260 ألف) توكن، ويمكن توسيعه عبر طرق استطراد مثل YaRN إلى مليون توكن، ممّا يمكّنه من التعامل مع مستودعات شفرة ضخمة ومهام برمجية معقّدة. صُمم Qwen3-Coder لسير عمل ترميز يعتمد على الوكلاء؛ فهو لا يولّد الشفرة فحسب، بل يتفاعل بشكلٍ مستقل مع أدوات وبيئات التطوير لحل مشكلات برمجية معقّدة. في اختبارات معيارية متعددة لمهام التكويد والوكالة، حقق النموذج مستوى متقدمًا بين النماذج مفتوحة المصدر، ويمكن أن ينافس نماذج رائدة مثل Claude Sonnet 4."
+  },
   "Qwen2-72B-Instruct": {
     "description": "Qwen2 هو أحدث سلسلة من نموذج Qwen، ويدعم سياقًا يصل إلى 128 ألف، مقارنةً بأفضل النماذج مفتوحة المصدر الحالية، يتفوق Qwen2-72B بشكل ملحوظ في فهم اللغة الطبيعية والمعرفة والترميز والرياضيات والقدرات متعددة اللغات."
   },
@@ -368,6 +398,9 @@
   "Qwen3-235B": {
     "description": "Qwen3-235B-A22B هو نموذج MoE (نموذج خبير مختلط) يقدم \"وضع الاستدلال المختلط\"، ويدعم المستخدمين في التبديل السلس بين \"وضع التفكير\" و\"وضع عدم التفكير\". يدعم فهم واستدلال 119 لغة ولهجة، ويتميز بقدرات قوية على استدعاء الأدوات. في اختبارات الأداء الشاملة، والبرمجة والرياضيات، واللغات المتعددة، والمعرفة والاستدلال، ينافس هذا النموذج النماذج الرائدة في السوق مثل DeepSeek R1، OpenAI o1، o3-mini، Grok 3، وGoogle Gemini 2.5 Pro."
   },
+  "Qwen3-235B-A22B-Instruct-2507-FP8": {
+    "description": "Qwen3 235B A22B Instruct 2507: نموذج موجه للاستدلال المتقدم وتحسين أوامر الحوار، يعتمد على بنية خبراء مختلطة للحفاظ على كفاءة الاستدلال مع عدد كبير من المعاملات."
+  },
   "Qwen3-32B": {
     "description": "Qwen3-32B هو نموذج كثيف (Dense Model) يقدم \"وضع الاستدلال المختلط\"، ويدعم التبديل السلس بين \"وضع التفكير\" و\"وضع عدم التفكير\". بفضل تحسينات في بنية النموذج، وزيادة بيانات التدريب، وأساليب تدريب أكثر فعالية، يقدم أداءً يعادل تقريبًا Qwen2.5-72B."
   },
@@ -824,6 +857,9 @@
   "deepseek-ai/DeepSeek-V3": {
     "description": "DeepSeek-V3 هو نموذج لغوي مختلط الخبراء (MoE) يحتوي على 6710 مليار معلمة، يستخدم انتباه متعدد الرؤوس (MLA) وبنية DeepSeekMoE، ويجمع بين استراتيجية توازن الحمل بدون خسارة مساعدة، مما يحسن كفاءة الاستدلال والتدريب. من خلال التدريب المسبق على 14.8 تريليون توكن عالي الجودة، وإجراء تعديلات إشرافية وتعلم معزز، يتفوق DeepSeek-V3 في الأداء على نماذج المصدر المفتوح الأخرى، ويقترب من النماذج المغلقة الرائدة."
   },
+  "deepseek-ai/DeepSeek-V3.1": {
+    "description": "DeepSeek-V3.1 هو نموذج لغة كبير بنمط هجين أصدرته DeepSeek AI، وقد شهد ترقيات مهمة متعددة مقارنة بالإصدارات السابقة. من الابتكارات الرئيسية في هذا النموذج دمج \"وضع التفكير\" و\"وضع عدم التفكير\" في نموذج واحد، حيث يمكن للمستخدمين التبديل بينهما بسهولة عبر تعديل قالب المحادثة لتلبية متطلبات المهام المختلفة. من خلال تحسينات ما بعد التدريب المخصصة، تم تعزيز أداء V3.1 في استدعاء الأدوات ومهام الوكيل بشكل ملحوظ، مما يمكنه من دعم أدوات البحث الخارجية وتنفيذ مهام معقدة متعددة الخطوات بشكل أفضل. يعتمد النموذج على DeepSeek-V3.1-Base مع تدريب إضافي، حيث تم توسيع حجم بيانات التدريب بشكل كبير عبر طريقة التوسيع النصي الطويل على مرحلتين، مما يحسن أدائه في معالجة المستندات الطويلة والرموز البرمجية الطويلة. كنموذج مفتوح المصدر، يظهر DeepSeek-V3.1 قدرة تنافسية مع أفضل النماذج المغلقة في مجالات الترميز والرياضيات والاستدلال، وبفضل هيكله المختلط للخبراء (MoE)، يحافظ على سعة نموذج ضخمة مع تقليل تكلفة الاستدلال بفعالية."
+  },
   "deepseek-ai/deepseek-llm-67b-chat": {
     "description": "DeepSeek 67B هو نموذج متقدم تم تدريبه للحوار المعقد."
   },
@@ -894,7 +930,7 @@
     "description": "DeepSeek R1 النسخة الكاملة، تحتوي على 671 مليار معلمة، تدعم البحث المتصل في الوقت الحقيقي، وتتمتع بقدرات فهم وتوليد أقوى."
   },
   "deepseek-reasoner": {
-    "description": "نموذج الاستدلال الذي أطلقته DeepSeek. قبل تقديم الإجابة النهائية، يقوم النموذج أولاً بإخراج سلسلة من المحتوى الفكري لتحسين دقة الإجابة النهائية."
+    "description": "وضع التفكير في DeepSeek V3.1. قبل إخراج الإجابة النهائية، يقوم النموذج أولاً بإخراج سلسلة من التفكير لتحسين دقة الإجابة النهائية."
   },
   "deepseek-v2": {
     "description": "DeepSeek V2 هو نموذج لغوي فعال من نوع Mixture-of-Experts، مناسب لاحتياجات المعالجة الاقتصادية."
@@ -908,12 +944,18 @@
   "deepseek-v3-0324": {
     "description": "DeepSeek-V3-0324 هو نموذج MoE يحتوي على 671 مليار معلمة، ويتميز بقدرات بارزة في البرمجة والتقنية، وفهم السياق ومعالجة النصوص الطويلة."
   },
+  "deepseek-v3.1": {
+    "description": "DeepSeek-V3.1 هو نموذج استدلال هجين جديد أطلقته DeepSeek، يدعم وضعين للاستدلال: التفكير وعدم التفكير، مع كفاءة تفكير أعلى مقارنة بـ DeepSeek-R1-0528. بعد تحسين ما بعد التدريب، تم تعزيز استخدام أدوات الوكيل وأداء مهام الوكيل بشكل كبير. يدعم نافذة سياق تصل إلى 128 ألف، وطول إخراج يصل إلى 64 ألف رمز."
+  },
   "deepseek/deepseek-chat-v3-0324": {
     "description": "DeepSeek V3 هو نموذج مختلط خبير يحتوي على 685B من المعلمات، وهو أحدث إصدار من سلسلة نماذج الدردشة الرائدة لفريق DeepSeek.\n\nيستفيد من نموذج [DeepSeek V3](/deepseek/deepseek-chat-v3) ويظهر أداءً ممتازًا في مجموعة متنوعة من المهام."
   },
   "deepseek/deepseek-chat-v3-0324:free": {
     "description": "DeepSeek V3 هو نموذج مختلط خبير يحتوي على 685B من المعلمات، وهو أحدث إصدار من سلسلة نماذج الدردشة الرائدة لفريق DeepSeek.\n\nيستفيد من نموذج [DeepSeek V3](/deepseek/deepseek-chat-v3) ويظهر أداءً ممتازًا في مجموعة متنوعة من المهام."
   },
+  "deepseek/deepseek-chat-v3.1": {
+    "description": "DeepSeek-V3.1 هو نموذج استدلال هجين كبير يدعم سياق طويل يصل إلى 128K وتبديل أوضاع فعال، ويحقق أداءً وسرعة ممتازة في استدعاء الأدوات، وتوليد الأكواد، والمهام الاستدلالية المعقدة."
+  },
   "deepseek/deepseek-r1": {
     "description": "DeepSeek-R1 يعزز بشكل كبير من قدرة النموذج على الاستدلال في ظل وجود بيانات محدودة جدًا. قبل تقديم الإجابة النهائية، يقوم النموذج أولاً بإخراج سلسلة من التفكير لتحسين دقة الإجابة النهائية."
   },
@@ -1103,12 +1145,27 @@
   "flux-dev": {
     "description": "FLUX.1 [dev] هو نموذج مفتوح المصدر للأوزان المكررة موجه للتطبيقات غير التجارية. يحافظ على جودة الصور وقدرة اتباع التعليمات مماثلة لإصدار FLUX الاحترافي، مع كفاءة تشغيل أعلى. مقارنة بالنماذج القياسية ذات الحجم المماثل، يستخدم الموارد بشكل أكثر فعالية."
   },
+  "flux-kontext-max": {
+    "description": "توليد وتحرير الصور السياقية بأحدث التقنيات — يجمع بين النص والصور لتحقيق نتائج دقيقة ومتسقة."
+  },
+  "flux-kontext-pro": {
+    "description": "توليد وتحرير الصور السياقية بأحدث التقنيات — يجمع بين النص والصورة للحصول على نتائج دقيقة ومتسقة."
+  },
   "flux-kontext/dev": {
     "description": "نموذج FLUX.1 مخصص لمهام تحرير الصور، يدعم إدخال النصوص والصور."
   },
   "flux-merged": {
     "description": "نموذج FLUX.1-merged يجمع بين ميزات العمق التي استكشفتها نسخة \"DEV\" أثناء التطوير ومزايا التنفيذ السريع التي تمثلها نسخة \"Schnell\". من خلال هذا الدمج، يعزز FLUX.1-merged حدود أداء النموذج ويوسع نطاق تطبيقاته."
   },
+  "flux-pro": {
+    "description": "نموذج توليد صور بالذكاء الاصطناعي من الطراز الأول للاستخدام التجاري — جودة صور لا تضاهى وقدرة عالية على إنتاج مخرجات متنوعة."
+  },
+  "flux-pro-1.1": {
+    "description": "نموذج متقدم احترافي لتوليد الصور بالذكاء الاصطناعي — يوفر جودة صور استثنائية وقدرة دقيقة على الالتزام بالمطالبات."
+  },
+  "flux-pro-1.1-ultra": {
+    "description": "توليد صور بالذكاء الاصطناعي بدقة فائقة — يدعم إخراج يصل إلى 4 ميجابكسل ويولد صورًا فائقة الوضوح خلال 10 ثوانٍ."
+  },
   "flux-pro/kontext": {
     "description": "FLUX.1 Kontext [pro] قادر على معالجة النصوص والصور المرجعية كمدخلات، مما يتيح تحريرًا محليًا مستهدفًا وتحولات معقدة للمشهد الكلي بسلاسة."
   },
@@ -1193,6 +1250,12 @@
   "gemini-2.5-flash": {
     "description": "Gemini 2.5 Flash هو نموذج Google الأكثر فعالية من حيث التكلفة، ويوفر وظائف شاملة."
   },
+  "gemini-2.5-flash-image-preview": {
+    "description": "Nano Banana هو أحدث وأسرع وأكثر نموذج متعدد الوسائط أصلي كفاءة من Google، يتيح لك إنشاء وتحرير الصور من خلال المحادثة."
+  },
+  "gemini-2.5-flash-image-preview:image": {
+    "description": "Nano Banana هو أحدث وأسرع وأكثر نموذج متعدد الوسائط أصلي كفاءة من Google، يتيح لك إنشاء وتحرير الصور من خلال المحادثة."
+  },
   "gemini-2.5-flash-lite": {
     "description": "Gemini 2.5 Flash-Lite هو أصغر وأفضل نموذج من حيث التكلفة من Google، مصمم للاستخدام على نطاق واسع."
   },
@@ -1295,6 +1358,9 @@
   "glm-4.5-x": {
     "description": "نسخة فائقة السرعة من GLM-4.5، تجمع بين أداء قوي وسرعة توليد تصل إلى 100 رمز في الثانية."
   },
+  "glm-4.5v": {
+    "description": "نموذج استدلال بصري من الجيل الجديد لشركة Zhipu مبني على بنية MOE، بإجمالي 106 مليار معامل و12 مليار معامل نشط، وقد بلغ مستوى الأداء الأعلى (SOTA) بين نماذج التعدد الوسائط مفتوحة المصدر المماثلة على مستوى العالم في عدة اختبارات معيارية، ويغطي مهامًا شائعة مثل فهم الصور والفيديو والمستندات وواجهات المستخدم الرسومية (GUI)."
+  },
   "glm-4v": {
     "description": "GLM-4V يوفر قدرات قوية في فهم الصور والاستدلال، ويدعم مجموعة متنوعة من المهام البصرية."
   },
@@ -1331,6 +1397,12 @@
   "google/gemini-2.5-flash": {
     "description": "Gemini 2.5 Flash هو النموذج الرئيسي الأكثر تقدمًا من Google، مصمم خصيصًا للمهام المتقدمة في الاستدلال، الترميز، الرياضيات والعلوم. يحتوي على قدرة مدمجة على \"التفكير\"، مما يمكنه من تقديم استجابات بدقة أعلى ومعالجة سياقية أكثر تفصيلاً.\n\nملاحظة: يحتوي هذا النموذج على نسختين: نسخة التفكير ونسخة غير التفكير. تختلف تكلفة الإخراج بشكل ملحوظ بناءً على تفعيل قدرة التفكير. إذا اخترت النسخة القياسية (بدون لاحقة \":thinking\"), سيتجنب النموذج بوضوح توليد رموز التفكير.\n\nلاستغلال قدرة التفكير واستلام رموز التفكير، يجب عليك اختيار النسخة \":thinking\"، والتي ستؤدي إلى تكلفة إخراج أعلى للتفكير.\n\nبالإضافة إلى ذلك، يمكن تكوين Gemini 2.5 Flash من خلال معلمة \"الحد الأقصى لعدد رموز الاستدلال\" كما هو موضح في الوثائق (https://openrouter.ai/docs/use-cases/reasoning-tokens#max-tokens-for-reasoning)."
   },
+  "google/gemini-2.5-flash-image-preview": {
+    "description": "نموذج تجريبي Gemini 2.5 Flash، يدعم توليد الصور."
+  },
+  "google/gemini-2.5-flash-image-preview:free": {
+    "description": "نموذج تجريبي Gemini 2.5 Flash، يدعم توليد الصور."
+  },
   "google/gemini-2.5-flash-preview": {
     "description": "Gemini 2.5 Flash هو النموذج الرائد الأكثر تقدمًا من Google، مصمم للاستدلال المتقدم، الترميز، المهام الرياضية والعلمية. يحتوي على قدرة \"التفكير\" المدمجة، مما يمكّنه من تقديم استجابات بدقة أعلى ومعالجة سياقات أكثر تفصيلاً.\n\nملاحظة: يحتوي هذا النموذج على نوعين: التفكير وغير التفكير. تختلف تسعير الإخراج بشكل ملحوظ بناءً على ما إذا كانت قدرة التفكير مفعلة. إذا اخترت النوع القياسي (بدون لاحقة \" :thinking \")، سيتجنب النموذج بشكل صريح توليد رموز التفكير.\n\nلاستغلال قدرة التفكير واستقبال رموز التفكير، يجب عليك اختيار النوع \" :thinking \"، مما سيؤدي إلى تسعير إخراج تفكير أعلى.\n\nبالإضافة إلى ذلك، يمكن تكوين Gemini 2.5 Flash من خلال معلمة \"الحد الأقصى لعدد رموز الاستدلال\"، كما هو موضح في الوثائق (https://openrouter.ai/docs/use-cases/reasoning-tokens#max-tokens-for-reasoning)."
   },
@@ -1434,7 +1506,7 @@
     "description": "يوفر GPT-4.1 mini توازنًا بين الذكاء والسرعة والتكلفة، مما يجعله نموذجًا جذابًا للعديد من الاستخدامات."
   },
   "gpt-4.5-preview": {
-    "description": "نسخة المعاينة البحثية لـ GPT-4.5، وهي أكبر وأقوى نموذج GPT لدينا حتى الآن. تتمتع بمعرفة واسعة عن العالم وتفهم أفضل لنوايا المستخدم، مما يجعلها بارعة في المهام الإبداعية والتخطيط الذاتي. يمكن لـ GPT-4.5 قبول المدخلات النصية والصورية وتوليد مخرجات نصية (بما في ذلك المخرجات الهيكلية). تدعم ميزات المطورين الأساسية مثل استدعاء الدوال، وواجهة برمجة التطبيقات الجماعية، والمخرجات المتدفقة. تتألق GPT-4.5 بشكل خاص في المهام التي تتطلب التفكير الإبداعي، والتفكير المفتوح، والحوار (مثل الكتابة، والتعلم، أو استكشاف أفكار جديدة). تاريخ انتهاء المعرفة هو أكتوبر 2023."
+    "description": "GPT-4.5-preview هو أحدث نموذج عام متعدد الأغراض، يتمتع بمعرفة عالمية عميقة وفهم محسن لنوايا المستخدم، ويتفوق في المهام الإبداعية وفي تخطيط الوكلاء. معارف هذا النموذج محدّثة حتى أكتوبر 2023."
   },
   "gpt-4o": {
     "description": "ChatGPT-4o هو نموذج ديناميكي يتم تحديثه في الوقت الحقيقي للحفاظ على أحدث إصدار. يجمع بين فهم اللغة القوي وقدرات التوليد، مما يجعله مناسبًا لمجموعة واسعة من التطبيقات، بما في ذلك خدمة العملاء والتعليم والدعم الفني."
@@ -1502,6 +1574,9 @@
   "gpt-oss": {
     "description": "GPT-OSS 20B هو نموذج لغة كبير مفتوح المصدر أصدرته OpenAI، يستخدم تقنية التكميم MXFP4، ومناسب للتشغيل على وحدات معالجة الرسومات الاستهلاكية المتقدمة أو أجهزة Mac بمعالج Apple Silicon. يتميز هذا النموذج بأداء ممتاز في توليد المحادثات، وكتابة الأكواد، ومهام الاستدلال، ويدعم استدعاء الدوال واستخدام الأدوات."
   },
+  "gpt-oss-120b": {
+    "description": "GPT-OSS-120B MXFP4: هيكل Transformer محسّن بالكمية، يحافظ على أداء قوي حتى في ظل محدودية الموارد."
+  },
   "gpt-oss:120b": {
     "description": "GPT-OSS 120B هو نموذج لغة كبير مفتوح المصدر أصدرته OpenAI، يستخدم تقنية التكميم MXFP4، ويعتبر نموذجًا رائدًا. يتطلب تشغيله بيئة متعددة وحدات معالجة الرسومات أو محطة عمل عالية الأداء، ويتميز بأداء متفوق في الاستدلال المعقد، وتوليد الأكواد، ومعالجة اللغات المتعددة، ويدعم استدعاء الدوال المتقدمة وتكامل الأدوات."
   },
@@ -1529,6 +1604,9 @@
   "grok-4": {
     "description": "نموذجنا الرائد الأحدث والأقوى، يتميز بأداء ممتاز في معالجة اللغة الطبيعية، الحسابات الرياضية، والاستدلال — إنه لاعب شامل مثالي."
   },
+  "grok-code-fast-1": {
+    "description": "نحن سعداء بإطلاق grok-code-fast-1، وهو نموذج استدلال سريع وفعال من حيث التكلفة، يتميز بأداء ممتاز في ترميز الوكلاء."
+  },
   "gryphe/mythomax-l2-13b": {
     "description": "MythoMax l2 13B هو نموذج لغوي يجمع بين الإبداع والذكاء من خلال دمج عدة نماذج رائدة."
   },
@@ -1637,9 +1715,18 @@
   "image-01-live": {
     "description": "نموذج توليد صور يقدم تفاصيل دقيقة، يدعم توليد الصور من النصوص مع إمكانية ضبط الأسلوب الفني."
   },
+  "imagen-4.0-fast-generate-001": {
+    "description": "Imagen، سلسلة نماذج تحويل النص إلى صورة من الجيل الرابع، الإصدار السريع"
+  },
+  "imagen-4.0-generate-001": {
+    "description": "سلسلة نماذج Imagen لتحويل النص إلى صورة من الجيل الرابع"
+  },
   "imagen-4.0-generate-preview-06-06": {
     "description": "سلسلة نموذج Imagen للجيل الرابع لتحويل النص إلى صورة"
   },
+  "imagen-4.0-ultra-generate-001": {
+    "description": "سلسلة نماذج Imagen لتحويل النص إلى صورة من الجيل الرابع — إصدار Ultra"
+  },
   "imagen-4.0-ultra-generate-preview-06-06": {
     "description": "نسخة ألترا من سلسلة نموذج Imagen للجيل الرابع لتحويل النص إلى صورة"
   },
@@ -1679,6 +1766,9 @@
   "kimi-k2-0711-preview": {
     "description": "kimi-k2 هو نموذج أساسي بمعمارية MoE يتمتع بقدرات فائقة في البرمجة والوكيل، مع إجمالي 1 تريليون معلمة و32 مليار معلمة مفعلة. في اختبارات الأداء الأساسية في مجالات المعرفة العامة، البرمجة، الرياضيات، والوكيل، يتفوق نموذج K2 على النماذج المفتوحة المصدر الرئيسية الأخرى."
   },
+  "kimi-k2-turbo-preview": {
+    "description": "kimi-k2 هو نموذج أساسي بمعمارية MoE يتمتع بقدرات قوية للغاية في البرمجة وقدرات الوكيل (Agent)، بإجمالي معلمات يبلغ 1 تريليون والمعلمات المُفعَّلة 32 مليار. في اختبارات الأداء المعيارية للفئات الرئيسية مثل الاستدلال المعرفي العام والبرمجة والرياضيات والوكلاء (Agent)، تفوق أداء نموذج K2 على النماذج المفتوحة المصدر السائدة الأخرى."
+  },
   "kimi-latest": {
     "description": "يستخدم منتج كيمي المساعد الذكي أحدث نموذج كبير من كيمي، وقد يحتوي على ميزات لم تستقر بعد. يدعم فهم الصور، وسيختار تلقائيًا نموذج 8k/32k/128k كنموذج للتسعير بناءً على طول سياق الطلب."
   },
@@ -1763,6 +1853,9 @@
   "llava:34b": {
     "description": "LLaVA هو نموذج متعدد الوسائط يجمع بين مشفرات بصرية وVicuna، يستخدم لفهم بصري ولغوي قوي."
   },
+  "magistral-medium-latest": {
+    "description": "Magistral Medium 1.1 هو نموذج استدلال رائد أطلقته Mistral AI في يوليو 2025."
+  },
   "mathstral": {
     "description": "MathΣtral مصمم للبحث العلمي والاستدلال الرياضي، يوفر قدرة حسابية فعالة وتفسير النتائج."
   },
@@ -2094,7 +2187,7 @@
     "description": "o1-mini هو نموذج استدلال سريع وفعال من حيث التكلفة مصمم لتطبيقات البرمجة والرياضيات والعلوم. يحتوي هذا النموذج على 128K من السياق وتاريخ انتهاء المعرفة في أكتوبر 2023."
   },
   "o1-preview": {
-    "description": "o1 هو نموذج استدلال جديد من OpenAI، مناسب للمهام المعقدة التي تتطلب معرفة عامة واسعة. يحتوي هذا النموذج على 128K من السياق وتاريخ انتهاء المعرفة في أكتوبر 2023."
+    "description": "يركّز على الاستدلال المتقدّم وحل المشكلات المعقّدة، بما في ذلك المهام الرياضية والعلمية. مناسب للغاية للتطبيقات التي تتطلّب فهماً عميقاً للسياق وسير عمل مستقل."
   },
   "o1-pro": {
     "description": "نماذج سلسلة o1 مدربة بالتعلم المعزز، قادرة على التفكير قبل الإجابة وتنفيذ مهام استدلال معقدة. يستخدم نموذج o1-pro موارد حسابية أكبر للتفكير الأعمق، مما يضمن تقديم إجابات ذات جودة أعلى باستمرار."
@@ -2213,8 +2306,14 @@
   "qwen-coder-turbo-latest": {
     "description": "نموذج Qwen للبرمجة."
   },
+  "qwen-flash": {
+    "description": "سلسلة Qwen-Flash هي الأسرع والأقل تكلفة بشكل كبير، ومناسبة للمهام البسيطة."
+  },
   "qwen-image": {
-    "description": "نموذج قوي من فريق Qwen لتوليد الصور الخام، يتميز بقدرة مميزة على توليد النصوص الصينية وأنماط بصرية متنوعة للصور."
+    "description": "Qwen-Image هي نموذج عام لتوليد الصور يدعم أنماطًا فنية متعددة، ويتميز بقدرته على عرض النصوص المعقدة، خصوصًا النصوص بالصينية والإنجليزية. يدعم النموذج تخطيطات متعددة الأسطر، وتوليد نص على مستوى الفقرات، وتمثيل التفاصيل الدقيقة، مما يتيح إنشاء تصميمات معقدة تمزج بين النص والصورة."
+  },
+  "qwen-image-edit": {
+    "description": "أصدر فريق Qwen نموذجًا احترافيًا لتحرير الصور يدعم التحرير الدلالي وتحرير المظهر، ويستطيع تحرير النصوص بالصينية والإنجليزية بدقة، وتحقيق تحويلات النمط وتدوير الكائنات، وغيرها من عمليات تحرير الصور عالية الجودة."
   },
   "qwen-long": {
     "description": "نموذج Qwen العملاق للغة، يدعم سياقات نصية طويلة، بالإضافة إلى وظائف الحوار المستندة إلى الوثائق الطويلة والعديد من الوثائق."
@@ -2241,7 +2340,7 @@
     "description": "نموذج لغة ضخم من توغي، نسخة معززة، يدعم إدخال لغات مختلفة مثل الصينية والإنجليزية."
   },
   "qwen-turbo": {
-    "description": "نموذج لغة ضخم من توغي، يدعم إدخال لغات مختلفة مثل الصينية والإنجليزية."
+    "description": "لن يتم تحديث 通义千问 Turbo مستقبلًا؛ يُنصح باستبداله بـ 通义千问 Flash. 通义千问 هو نموذج لغوي ضخم جدًا ويدعم إدخال لغات مختلفة مثل الصينية والإنجليزية."
   },
   "qwen-vl-chat-v1": {
     "description": "نموذج Qwen العملاق للغة البصرية يدعم طرق تفاعل مرنة، بما في ذلك الصور المتعددة، والأسئلة والأجوبة المتعددة، والإبداع."
@@ -2558,9 +2657,15 @@
   "step-2x-large": {
     "description": "نموذج الجيل الجديد من Step Star، يركز على مهام توليد الصور، قادر على توليد صور عالية الجودة بناءً على الأوصاف النصية المقدمة من المستخدم. يتميز النموذج الجديد بجودة صور أكثر واقعية وقدرات أفضل في توليد النصوص الصينية والإنجليزية."
   },
+  "step-3": {
+    "description": "يمتلك هذا النموذج قدرة فائقة على الإدراك البصري وعلى الاستدلال المعقّد. يمكنه إنجاز فهم دقيق للمعارف المعقّدة عبر مجالات متعدّدة، وإجراء تحليلات تقاطعية بين المعلومات الرياضية والبصرية، وكذلك معالجة مختلف مسائل التحليل البصري في الحياة اليومية."
+  },
   "step-r1-v-mini": {
     "description": "هذا النموذج هو نموذج استدلال كبير يتمتع بقدرة قوية على فهم الصور، يمكنه معالجة المعلومات النصية والصورية، ويخرج نصوصًا بعد تفكير عميق. يظهر هذا النموذج أداءً بارزًا في مجال الاستدلال البصري، كما يمتلك قدرات رياضية، برمجية، ونصية من الدرجة الأولى. طول السياق هو 100k."
   },
+  "stepfun-ai/step3": {
+    "description": "Step3 هو نموذج استدلال متعدد الوسائط متقدم أصدرته شركة 阶跃星辰 (StepFun). بُني على بنية مزيج الخبراء (MoE) التي تضم 321 مليار معلمة إجمالية و38 مليار معلمة تنشيط. صُمم النموذج بنهج من الطرف إلى الطرف ليقلل تكلفة فك الترميز، مع تقديم أداء رائد في الاستدلال البصري-اللغوي. من خلال التصميم التعاوني لآلية انتباه تفكيك متعدد المصفوفات (MFA) وفصل الانتباه عن شبكة التغذية الأمامية (AFD)، يحافظ Step3 على كفاءة ممتازة على كل من المسرعات الرائدة والمسرعات منخفضة التكلفة. في مرحلة ما قبل التدريب عالج Step3 أكثر من 20 تريليون توكن نصي و4 تريليون توكن مختلط نص-صورة، مغطياً أكثر من عشر لغات. حقق النموذج أداءً متقدماً بين نماذج المصدر المفتوح في عدة معايير قياسية تشمل الرياضيات والبرمجة والمهام متعددة الوسائط."
+  },
   "taichu_llm": {
     "description": "نموذج اللغة الكبير TaiChu يتمتع بقدرات قوية في فهم اللغة، بالإضافة إلى إنشاء النصوص، والإجابة على الأسئلة، وبرمجة الأكواد، والحسابات الرياضية، والاستدلال المنطقي، وتحليل المشاعر، وتلخيص النصوص. يجمع بشكل مبتكر بين التدريب المسبق على البيانات الضخمة والمعرفة الغنية من مصادر متعددة، من خلال تحسين تقنيات الخوارزميات باستمرار واستيعاب المعرفة الجديدة من البيانات النصية الضخمة، مما يحقق تطورًا مستمرًا في أداء النموذج. يوفر للمستخدمين معلومات وخدمات أكثر سهولة وتجربة أكثر ذكاءً."
   },
@@ -2588,18 +2693,9 @@
   "thudm/glm-4-9b-chat": {
     "description": "الإصدار المفتوح من الجيل الأحدث من نموذج GLM-4 الذي أطلقته Zhizhu AI."
   },
-  "thudm/glm-4-9b:free": {
-    "description": "GLM-4-9B-0414 هو نموذج لغوي يحتوي على 9 مليار معلمة من سلسلة GLM-4 التي تم تطويرها بواسطة THUDM. يستخدم GLM-4-9B-0414 نفس استراتيجيات تعزيز التعلم والتوافق المستخدمة في النموذج المقابل الأكبر 32B، مما يحقق أداءً عاليًا بالنسبة لحجمه، مما يجعله مناسبًا للنشر في البيئات المحدودة الموارد التي لا تزال تتطلب قدرات قوية في فهم اللغة وتوليدها."
-  },
   "thudm/glm-z1-32b": {
     "description": "GLM-Z1-32B-0414 هو نسخة محسنة من GLM-4-32B، مصممة لحل المشكلات المعقدة في الرياضيات العميقة، المنطق، والشيفرات. يستخدم التعلم المعزز الموسع (المخصص للمهام والمبني على تفضيلات عامة) لتحسين الأداء في المهام المعقدة متعددة الخطوات. مقارنةً بنموذج GLM-4-32B الأساسي، زادت Z1 بشكل ملحوظ من قدرات الاستدلال الهيكلي والمجالات الرسمية.\n\nيدعم هذا النموذج تنفيذ خطوات \"التفكير\" من خلال هندسة التلميحات، ويقدم اتساقًا محسنًا للإخراج الطويل. تم تحسينه لعمليات سير العمل الخاصة بالوكيل، ويدعم السياقات الطويلة (عبر YaRN)، واستدعاءات أدوات JSON، وتكوينات أخذ العينات الدقيقة للاستدلال المستقر. مثالي للحالات التي تتطلب تفكيرًا عميقًا، استدلالًا متعدد الخطوات، أو استنتاجات رسمية."
   },
-  "thudm/glm-z1-32b:free": {
-    "description": "GLM-Z1-32B-0414 هو نسخة محسنة من GLM-4-32B، مصممة لحل المشكلات المعقدة في الرياضيات العميقة، المنطق، والشيفرات. يستخدم التعلم المعزز الموسع (المخصص للمهام والمبني على تفضيلات عامة) لتحسين الأداء في المهام المعقدة متعددة الخطوات. مقارنةً بنموذج GLM-4-32B الأساسي، زادت Z1 بشكل ملحوظ من قدرات الاستدلال الهيكلي والمجالات الرسمية.\n\nيدعم هذا النموذج تنفيذ خطوات \"التفكير\" من خلال هندسة التلميحات، ويقدم اتساقًا محسنًا للإخراج الطويل. تم تحسينه لعمليات سير العمل الخاصة بالوكيل، ويدعم السياقات الطويلة (عبر YaRN)، واستدعاءات أدوات JSON، وتكوينات أخذ العينات الدقيقة للاستدلال المستقر. مثالي للحالات التي تتطلب تفكيرًا عميقًا، استدلالًا متعدد الخطوات، أو استنتاجات رسمية."
-  },
-  "thudm/glm-z1-9b:free": {
-    "description": "GLM-Z1-9B-0414 هو نموذج لغوي يحتوي على 9 مليار معلمة من سلسلة GLM-4 التي تم تطويرها بواسطة THUDM. يستخدم تقنيات تم تطبيقها في الأصل على نموذج GLM-Z1 الأكبر، بما في ذلك تعزيز التعلم الموسع، والتوافق القائم على الترتيب الثنائي، والتدريب على المهام التي تتطلب استدلالًا مكثفًا مثل الرياضيات، والترميز، والمنطق. على الرغم من حجمه الأصغر، إلا أنه يظهر أداءً قويًا في المهام العامة للاستدلال، ويتفوق على العديد من النماذج مفتوحة المصدر في مستوى وزنه."
-  },
   "thudm/glm-z1-rumination-32b": {
     "description": "THUDM: GLM Z1 Rumination 32B هو نموذج استدلال عميق يحتوي على 32 مليار معلمة من سلسلة GLM-4-Z1، تم تحسينه للمهام المعقدة والمفتوحة التي تتطلب تفكيرًا طويل الأمد. يعتمد على glm-4-32b-0414، ويضيف مراحل تعزيز التعلم الإضافية واستراتيجيات التوافق متعددة المراحل، ويقدم قدرة \"التفكير\" المصممة لمحاكاة معالجة الإدراك الموسع. يشمل ذلك الاستدلال التكراري، والتحليل متعدد القفزات، وسير العمل المعزز بالأدوات مثل البحث، والاسترجاع، والتوليف المدرك للاقتباسات.\n\nيظهر هذا النموذج أداءً ممتازًا في الكتابة البحثية، والتحليل المقارن، والأسئلة المعقدة. يدعم استدعاء الوظائف المستخدمة في البحث والتنقل (مثل `search`، `click`، `open`، `finish`)، مما يسمح باستخدامه في أنابيب الوكلاء. يتم تشكيل سلوك التفكير من خلال مكافآت قائمة على القواعد وآلية اتخاذ القرار المتأخرة، ويتم قياسه باستخدام أطر بحث عميقة مثل كومة التوافق الداخلية لـ OpenAI. هذا المتغير مناسب للسيناريوهات التي تتطلب عمقًا بدلاً من السرعة."
   },
@@ -2707,5 +2803,8 @@
   },
   "zai-org/GLM-4.5-Air": {
     "description": "GLM-4.5-Air هو نموذج أساسي مصمم لتطبيقات الوكلاء الذكية، يستخدم بنية Mixture-of-Experts (MoE). تم تحسينه بعمق في مجالات استدعاء الأدوات، تصفح الويب، هندسة البرمجيات، وبرمجة الواجهة الأمامية، ويدعم التكامل السلس مع وكلاء الكود مثل Claude Code وRoo Code. يستخدم وضع استدلال مختلط ليتكيف مع سيناريوهات الاستدلال المعقدة والاستخدام اليومي."
+  },
+  "zai-org/GLM-4.5V": {
+    "description": "GLM-4.5V هو نموذج لغوي بصري (VLM) من الجيل الأحدث صدر عن Zhipu AI (智谱 AI). بُني النموذج على نموذج النص الرائد GLM-4.5-Air الذي يحتوي على 106B من المعاملات الإجمالية و12B من معاملات التنشيط، ويعتمد على بنية الخبراء المختلطين (MoE) بهدف تحقيق أداء متميز بتكلفة استدلال أقل. من الناحية التقنية، يواصل GLM-4.5V نهج GLM-4.1V-Thinking ويقدّم ابتكارات مثل ترميز المواقع الدوراني ثلاثي الأبعاد (3D-RoPE)، مما عزّز بشكل ملحوظ قدرته على إدراك واستنتاج العلاقات المكانية ثلاثية الأبعاد. وبفضل تحسينات في مراحل ما قبل التدريب، والتعديل بالإشراف، والتعلّم المعزّز، أصبح النموذج قادراً على معالجة محتوى بصري متنوّع مثل الصور والفيديوهات والمستندات الطويلة، وقد حقق مستوى متقدماً ضمن أفضل نماذج المصدر المفتوح في 41 معياراً متعدد الوسائط منشوراً. بالإضافة إلى ذلك، أضاف النموذج مفتاح \"وضع التفكير\" الذي يتيح للمستخدمين التبديل بين الاستجابة السريعة والاستدلال العميق بحرية لتوازن أفضل بين الكفاءة والفعالية."
   }
 }

locales/ar/providers.json

@@ -11,6 +11,9 @@
   "aihubmix": {
     "description": "يوفر AiHubMix الوصول إلى نماذج الذكاء الاصطناعي المتعددة من خلال واجهة برمجة تطبيقات موحدة."
   },
+  "akashchat": {
+    "description": "آكاش هو سوق موارد سحابية بدون ترخيص، يتميز بأسعار تنافسية مقارنة بمزودي السحابة التقليديين."
+  },
   "anthropic": {
     "description": "Anthropic هي شركة تركز على أبحاث وتطوير الذكاء الاصطناعي، وتقدم مجموعة من نماذج اللغة المتقدمة، مثل Claude 3.5 Sonnet وClaude 3 Sonnet وClaude 3 Opus وClaude 3 Haiku. تحقق هذه النماذج توازنًا مثاليًا بين الذكاء والسرعة والتكلفة، وتناسب مجموعة متنوعة من سيناريوهات التطبيقات، من أحمال العمل على مستوى المؤسسات إلى الاستجابات السريعة. يعتبر Claude 3.5 Sonnet أحدث نماذجها، وقد أظهر أداءً ممتازًا في العديد من التقييمات مع الحفاظ على نسبة تكلفة فعالة."
   },
@@ -26,6 +29,9 @@
   "bedrock": {
     "description": "Bedrock هي خدمة تقدمها أمازون AWS، تركز على توفير نماذج لغة ورؤية متقدمة للذكاء الاصطناعي للشركات. تشمل عائلة نماذجها سلسلة Claude من Anthropic وسلسلة Llama 3.1 من Meta، وتغطي مجموعة من الخيارات من النماذج الخفيفة إلى عالية الأداء، وتدعم مهام مثل توليد النصوص، والحوار، ومعالجة الصور، مما يجعلها مناسبة لتطبيقات الشركات بمختلف أحجامها واحتياجاتها."
   },
+  "bfl": {
+    "description": "مختبر أبحاث رائد في مقدمة الذكاء الاصطناعي، يبني البنية التحتية البصرية للمستقبل."
+  },
   "cloudflare": {
     "description": "تشغيل نماذج التعلم الآلي المدفوعة بوحدات معالجة الرسوميات بدون خادم على شبكة Cloudflare العالمية."
   },
@@ -86,6 +92,12 @@
   "moonshot": {
     "description": "Moonshot هي منصة مفتوحة أطلقتها شركة Beijing Dark Side Technology Co.، Ltd، تقدم مجموعة متنوعة من نماذج معالجة اللغة الطبيعية، وتغطي مجالات واسعة، بما في ذلك ولكن لا تقتصر على إنشاء المحتوى، والبحث الأكاديمي، والتوصيات الذكية، والتشخيص الطبي، وتدعم معالجة النصوص الطويلة والمهام المعقدة."
   },
+  "nebius": {
+    "description": "نيبيوس توفر بنية تحتية عالية الأداء للمبتكرين في مجال الذكاء الاصطناعي حول العالم من خلال بناء مجموعات ضخمة من وحدات معالجة الرسومات ومنصة سحابية متكاملة رأسياً."
+  },
+  "newapi": {
+    "description": "منصة مفتوحة المصدر لتجميع وتحويل خدمات الذكاء الاصطناعي المتعددة بشكل موحد"
+  },
   "novita": {
     "description": "Novita AI هي منصة تقدم خدمات API لمجموعة متنوعة من نماذج اللغة الكبيرة وتوليد الصور بالذكاء الاصطناعي، مرنة وموثوقة وفعالة من حيث التكلفة. تدعم أحدث النماذج مفتوحة المصدر مثل Llama3 وMistral، وتوفر حلول API شاملة وسهلة الاستخدام وقابلة للتوسع تلقائيًا لتطوير تطبيقات الذكاء الاصطناعي، مما يجعلها مناسبة لنمو الشركات الناشئة في مجال الذكاء الاصطناعي."
   },

locales/ar/setting.json

@@ -237,6 +237,10 @@
       "title": "تحديد عدد الرسائل التاريخية",
       "unlimited": "غير محدود"
     },
+    "enableStreaming": {
+      "desc": "تمكين الإخراج المتدفق لعرض الاستجابات في الوقت الفعلي. عند التعطيل، يتم عرض الاستجابة الكاملة فقط.",
+      "title": "تمكين الإخراج المتدفق"
+    },
     "historyCount": {
       "desc": "عدد الرسائل التي يتم إرفاقها في كل طلب (تشمل الأسئلة والأجوبة الجديدة. يُحسب كل سؤال وجواب كرسالة واحدة)",
       "title": "عدد الرسائل المرفقة"

locales/bg-BG/chat.json

@@ -52,7 +52,11 @@
     "thinking": {
       "title": "Превключвател за дълбоко мислене"
     },
-    "title": "Разширени функции на модела"
+    "title": "Разширени функции на модела",
+    "urlContext": {
+      "desc": "Когато е включено, автоматично ще се анализират уеб връзки, за да се получи реалното съдържание на уеб страницата",
+      "title": "Извличане на съдържание от уеб връзки"
+    }
   },
   "history": {
     "title": "Асистентът ще запомни само последните {{count}} съобщения"
@@ -125,6 +129,7 @@
       "inputWriteCached": "Входящи кеширани данни",
       "output": "Изход",
       "outputAudio": "Аудио изход",
+      "outputImage": "Изходно изображение",
       "outputText": "Текстов изход",
       "outputTitle": "Детайли за изхода",
       "reasoning": "Дълбочинно разсъждение",
@@ -268,7 +273,8 @@
     "clientMode": {
       "actionFiletip": "Качване на файл",
       "actionTooltip": "Качване",
-      "disabled": "Текущият модел не поддържа визуално разпознаване и анализ на файлове, моля, превключете модела и опитайте отново"
+      "disabled": "Текущият модел не поддържа визуално разпознаване и анализ на файлове, моля, превключете модела и опитайте отново",
+      "visionNotSupported": "Текущият модел не поддържа визуално разпознаване, моля, превключете на друг модел, за да използвате тази функция"
     },
     "preview": {
       "prepareTasks": "Подготовка на парчета...",

locales/bg-BG/common.json

@@ -32,6 +32,7 @@
   "blog": "Продуктов блог",
   "branching": "Създаване на подтема",
   "branchingDisable": "Функцията „подтема“ е налична само в сървърната версия. Ако искате да използвате тази функция, моля, превключете на режим на сървърно разполагане или използвайте LobeChat Cloud.",
+  "branchingRequiresSavedTopic": "Текущата тема не е запазена, моля запазете я, за да използвате функцията за под-теми",
   "cancel": "Отказ",
   "changelog": "Дневник на промените",
   "clientDB": {
@@ -182,6 +183,13 @@
     "title": "Харесвате нашия продукт?"
   },
   "fullscreen": "Цял екран",
+  "geminiImageChineseWarning": {
+    "content": "Nano Banana може да не успее да генерира изображение при използване на китайски език. Препоръчваме използването на английски за по-добри резултати.",
+    "continueGenerate": "Продължи генерирането",
+    "continueSend": "Продължи изпращането",
+    "doNotShowAgain": "Не показвай отново",
+    "title": "Подсказка за въвеждане на китайски"
+  },
   "historyRange": "Диапазон на историята",
   "import": "Импортиране",
   "importData": "Импорт на данни",

locales/bg-BG/components.json

@@ -1,4 +1,9 @@
 {
+  "ArgsInput": {
+    "addArgument": "Добави параметър",
+    "argumentPlaceholder": "Параметър {{index}}",
+    "enterFirstArgument": "Въведете първия параметър..."
+  },
   "DragUpload": {
     "dragDesc": "Плъзнете файлове тук, поддържа качване на множество изображения.",
     "dragFileDesc": "Плъзнете изображения и файлове тук, поддържа качване на множество изображения и файлове.",
@@ -68,6 +73,18 @@
   "GoBack": {
     "back": "Назад"
   },
+  "HtmlPreview": {
+    "actions": {
+      "download": "Изтегляне",
+      "preview": "Преглед"
+    },
+    "iframeTitle": "HTML Преглед",
+    "mode": {
+      "code": "Код",
+      "preview": "Преглед"
+    },
+    "title": "HTML Преглед"
+  },
   "ImageUpload": {
     "actions": {
       "changeImage": "Кликнете, за да смените изображението",
@@ -125,6 +142,12 @@
     },
     "progress": {
       "uploadingWithCount": "{{completed}}/{{total}} качени"
+    },
+    "validation": {
+      "fileSizeExceeded": "Превишен допустим размер на файла",
+      "fileSizeExceededDetail": "{{fileName}} ({{actualSize}}) надвишава максимално допустимия размер {{maxSize}}",
+      "fileSizeExceededMultiple": "{{count}} файла надвишават максимално допустимия размер {{maxSize}}: {{fileList}}",
+      "imageCountExceeded": "Надвишен е допустимият брой изображения"
     }
   },
   "OllamaSetupGuide": {

locales/bg-BG/error.json

@@ -85,6 +85,17 @@
     "CreateMessageError": "Съжалявам, съобщението не можа да бъде изпратено успешно. Моля, копирайте съдържанието и го изпратете отново. След опресняване на страницата, това съобщение няма да бъде запазено.",
     "ExceededContextWindow": "Текущото съдържание на заявката надвишава дължината, която моделът може да обработи. Моля, намалете обема на съдържанието и опитайте отново.",
     "FreePlanLimit": "В момента сте потребител на безплатен план и не можете да използвате тази функционалност. Моля, надстройте до платен план, за да продължите да я използвате.",
+    "GoogleAIBlockReason": {
+      "BLOCKLIST": "Вашето съдържание съдържа забранени думи. Моля, проверете и коригирайте входа си и опитайте отново.",
+      "IMAGE_SAFETY": "Генерираното изображение е блокирано поради съображения за безопасност. Моля, опитайте да промените заявката за генериране на изображение.",
+      "LANGUAGE": "Езикът, който използвате, все още не се поддържа. Моля, опитайте да зададете въпроса на английски или на друг поддържан език.",
+      "OTHER": "Съдържанието е блокирано по неизвестна причина. Моля, опитайте да преформулирате заявката си.",
+      "PROHIBITED_CONTENT": "Вашата заявка може да съдържа забранено съдържание. Моля, коригирайте заявката, за да отговаря на правилата за ползване.",
+      "RECITATION": "Съдържанието ви е блокирано поради възможни проблеми с авторските права. Моля, опитайте да използвате оригинално съдържание или да преформулирате заявката си.",
+      "SAFETY": "Вашето съдържание е блокирано поради правила за безопасност. Моля, опитайте да промените заявката, като избягвате потенциално вредно или неподходящо съдържание.",
+      "SPII": "Вашето съдържание може да съдържа чувствителна лична информация. За да защитите поверителността, моля, премахнете съответната чувствителна информация и опитайте отново.",
+      "default": "Съдържанието е блокирано: {{blockReason}}。请调整您的请求内容后重试。"
+    },
     "InsufficientQuota": "Съжаляваме, квотата за този ключ е достигнала лимита. Моля, проверете баланса на акаунта си или увеличете квотата на ключа и опитайте отново.",
     "InvalidAccessCode": "Невалиден или празен код за достъп. Моля, въведете правилния код за достъп или добавете персонализиран API ключ.",
     "InvalidBedrockCredentials": "Удостоверяването на Bedrock е неуспешно. Моля, проверете AccessKeyId/SecretAccessKey и опитайте отново.",

locales/bg-BG/modelProvider.json

@@ -153,6 +153,28 @@
     "searchProviders": "Търсене на доставчици...",
     "sort": "Персонализирано сортиране"
   },
+  "newapi": {
+    "apiKey": {
+      "desc": "API ключът, предоставен от платформата New API",
+      "placeholder": "API ключ на New API",
+      "required": "API ключът е задължителен",
+      "title": "API ключ"
+    },
+    "apiUrl": {
+      "desc": "API адресът на услугата New API, обикновено трябва да включва /v1",
+      "title": "API адрес"
+    },
+    "enabled": {
+      "title": "Активиране на New API"
+    },
+    "models": {
+      "batchSelect": "Избор на множество модели ({{count}} броя)",
+      "fetch": "Вземане на списък с модели",
+      "selected": "Избрани модели",
+      "title": "Налични модели"
+    },
+    "title": "New API"
+  },
   "ollama": {
     "checker": {
       "desc": "Тестване дали адресът на прокси е попълнен правилно",
@@ -185,6 +207,10 @@
     }
   },
   "providerModels": {
+    "batchSelect": {
+      "selected": "Избрани {{count}} модела",
+      "title": "Множествен избор"
+    },
     "config": {
       "aesGcm": "Вашият ключ и адреса на прокси ще бъдат криптирани с <1>AES-GCM</1> алгоритъм",
       "apiKey": {

locales/bg-BG/models.json

@@ -74,9 +74,15 @@
   "DeepSeek-V3": {
     "description": "DeepSeek-V3 е MoE модел, разработен от компанията DeepSeek. DeepSeek-V3 постига резултати в множество оценки, които надминават други отворени модели като Qwen2.5-72B и Llama-3.1-405B, като по отношение на производителност е наравно с водещите затворени модели в света като GPT-4o и Claude-3.5-Sonnet."
   },
+  "DeepSeek-V3-1": {
+    "description": "DeepSeek V3.1: Следващо поколение модел за разсъждение, подобряващ сложното разсъждение и веригата на мислене, подходящ за задачи, изискващи задълбочен анализ."
+  },
   "DeepSeek-V3-Fast": {
     "description": "Доставчик на модела: платформа sophnet. DeepSeek V3 Fast е високоскоростната версия с висока TPS на DeepSeek V3 0324, с пълна точност без квантизация, с по-силни кодови и математически възможности и по-бърз отговор!"
   },
+  "DeepSeek-V3.1": {
+    "description": "DeepSeek-V3.1 е новият хибриден модел за разсъждение на DeepSeek, който поддържа два режима на разсъждение: мислене и немислене, с по-висока ефективност на мислене в сравнение с DeepSeek-R1-0528. След оптимизация чрез пост-тренировка, използването на агентски инструменти и изпълнението на задачи от интелигентни агенти са значително подобрени."
+  },
   "Doubao-lite-128k": {
     "description": "Doubao-lite предлага изключително бърза реакция и по-добро съотношение цена-качество, осигурявайки по-гъвкави опции за различни сценарии на клиентите. Поддържа разсъждения и финна настройка с контекстен прозорец от 128k."
   },
@@ -131,9 +137,15 @@
   "ERNIE-Speed-Pro-128K": {
     "description": "Най-новият модел на Baidu за големи езикови модели с висока производителност, разработен самостоятелно, с отлични общи способности, по-добри резултати в сравнение с ERNIE Speed, подходящ за основен модел за фина настройка, за по-добро справяне с конкретни проблеми, като същевременно предлага отлична производителност при извеждане."
   },
+  "FLUX-1.1-pro": {
+    "description": "FLUX.1.1 Pro"
+  },
   "FLUX.1-Kontext-dev": {
     "description": "FLUX.1-Kontext-dev е мултимоделен модел за генериране и редактиране на изображения, разработен от Black Forest Labs, базиран на архитектурата Rectified Flow Transformer с 12 милиарда параметри. Моделът е специализиран в генериране, реконструкция, подобряване и редактиране на изображения при зададени контекстуални условия. Той съчетава предимствата на контролираното генериране на дифузионни модели с контекстуалното моделиране на Transformer, поддържайки висококачествен изход и широко приложение в задачи като възстановяване, допълване и реконструкция на визуални сцени."
   },
+  "FLUX.1-Kontext-pro": {
+    "description": "FLUX.1 Kontext [pro]"
+  },
   "FLUX.1-dev": {
     "description": "FLUX.1-dev е отворен мултимодален езиков модел (Multimodal Language Model, MLLM), разработен от Black Forest Labs, оптимизиран за задачи с текст и изображения. Той интегрира разбиране и генериране на изображения и текст, базиран на напреднали големи езикови модели като Mistral-7B, с внимателно проектиран визуален енкодер и многостепенно фино настройване с инструкции, което позволява съвместна обработка на текст и изображения и сложни задачи за разсъждение."
   },
@@ -170,6 +182,9 @@
   "Llama-3.2-90B-Vision-Instruct\t": {
     "description": "Напреднали способности за визуално разсъждение, подходящи за приложения на агенти за визуално разбиране."
   },
+  "Meta-Llama-3-3-70B-Instruct": {
+    "description": "Llama 3.3 70B: Универсален трансформър модел, подходящ за диалог и задачи за генериране."
+  },
   "Meta-Llama-3.1-405B-Instruct": {
     "description": "Текстов модел с оптимизация за инструкции на Llama 3.1, проектиран за многоезични диалогови случаи, който показва отлични резултати на много налични отворени и затворени чат модели на общи индустриални бенчмаркове."
   },
@@ -188,6 +203,9 @@
   "Meta-Llama-3.3-70B-Instruct": {
     "description": "Llama 3.3 е най-напредналият многоезичен отворен голям езиков модел от серията Llama, който предлага производителност, сравнима с 405B моделите, на изключително ниска цена. Базиран на структурата Transformer и подобрен чрез супервизирано фино настройване (SFT) и обучение с човешка обратна връзка (RLHF) за повишаване на полезността и безопасността. Неговата версия с оптимизация за инструкции е специално проектирана за многоезични диалози и показва по-добри резултати от много от наличните отворени и затворени чат модели на множество индустриални бенчмаркове. Краен срок за знанията е декември 2023 г."
   },
+  "Meta-Llama-4-Maverick-17B-128E-Instruct-FP8": {
+    "description": "Llama 4 Maverick: Голям модел, базиран на Mixture-of-Experts, предлагащ ефективна стратегия за активиране на експерти за отлични резултати при разсъждение."
+  },
   "MiniMax-M1": {
     "description": "Изцяло ново самостоятелно разработено модел за разсъждение. Световен лидер: 80K вериги на мислене x 1M вход, с резултати, сравними с водещите модели в чужбина."
   },
@@ -257,6 +275,9 @@
   "Pro/deepseek-ai/DeepSeek-V3": {
     "description": "DeepSeek-V3 е модел на езика с 6710 милиарда параметри, който използва архитектура на смесени експерти (MoE) с много глави на потенциално внимание (MLA) и стратегия за баланс на натоварването без помощни загуби, оптимизираща производителността на инференцията и обучението. Чрез предварително обучение на 14.8 трилиона висококачествени токени и последващо супервизирано фино настройване и обучение с подсилване, DeepSeek-V3 надминава производителността на други отворени модели и е близо до водещите затворени модели."
   },
+  "Pro/deepseek-ai/DeepSeek-V3.1": {
+    "description": "DeepSeek-V3.1 е хибриден голям езиков модел, пуснат от DeepSeek AI, който включва множество важни подобрения спрямо предишните версии. Основната иновация на модела е интеграцията на „режим на мислене“ (Thinking Mode) и „режим без мислене“ (Non-thinking Mode), които потребителите могат гъвкаво да превключват чрез настройка на чат шаблони, за да отговарят на различни задачи. След специална пост-тренировка, V3.1 значително подобрява производителността при използване на инструменти и задачи на агенти, като по-добре поддържа външни търсачки и изпълнение на сложни многостъпкови задачи. Моделът е дообучен върху DeepSeek-V3.1-Base чрез двуфазен метод за разширяване на дълги текстове, което значително увеличава обема на тренировъчните данни и подобрява работата с дълги документи и кодове. Като отворен модел, DeepSeek-V3.1 демонстрира способности, сравними с водещи затворени модели в области като кодиране, математика и разсъждение, като същевременно с хибридната си експертна (MoE) архитектура поддържа голям капацитет на модела и ефективно намалява разходите за изчисления."
+  },
   "Pro/moonshotai/Kimi-K2-Instruct": {
     "description": "Kimi K2 е базов модел с MoE архитектура с изключителни кодови и агентски способности, с общо 1 трилион параметри и 32 милиарда активирани параметри. В бенчмаркове за общо знание, програмиране, математика и агентски задачи моделът K2 превъзхожда други водещи отворени модели."
   },
@@ -332,12 +353,21 @@
   "Qwen/Qwen3-30B-A3B-Instruct-2507": {
     "description": "Qwen3-30B-A3B-Instruct-2507 е обновена версия на Qwen3-30B-A3B в режим без мислене. Това е хибриден експертен (MoE) модел с общо 30,5 милиарда параметри и 3,3 милиарда активни параметри. Моделът е получил ключови подобрения в множество аспекти, включително значително подобрена способност за следване на инструкции, логическо разсъждение, разбиране на текст, математика, наука, кодиране и използване на инструменти. Освен това, той постига съществен напредък в покритието на дългоопашатите знания на многоезично ниво и по-добре се съгласува с предпочитанията на потребителите при субективни и отворени задачи, което позволява генериране на по-полезни отговори и по-висококачествен текст. Освен това, способността му за разбиране на дълги текстове е увеличена до 256K. Този модел поддържа само режим без мислене и в изхода му не се генерират тагове `<think></think>`."
   },
+  "Qwen/Qwen3-30B-A3B-Thinking-2507": {
+    "description": "Qwen3-30B-A3B-Thinking-2507 е най-новият мисловен модел от серията Qwen3, публикуван от екипа Tongyi Qianwen на Alibaba. Като хибриден модел от типа Mixture of Experts (MoE) с общо 30,5 милиарда параметри и 3,3 милиарда активни параметри, той е насочен към повишаване на възможностите за справяне със сложни задачи. Моделът показва значително подобрение в логическото разсъждение, математиката, естествените науки, програмирането и академичните бенчмаркове, изискващи човешка експертиза. В същото време общите му способности — следване на инструкции, използване на инструменти, генериране на текст и съгласуване с човешките предпочитания — също са значително подсилени. Моделът поддържа нативно дълъг контекст до 256K и може да бъде разширен до 1 000 000 токена. Тази версия е специално проектирана за мисловен режим, с цел да решава изключително сложни задачи чрез подробни стъпкови разсъждения; агентните ѝ възможности също се представят отлично."
+  },
   "Qwen/Qwen3-32B": {
     "description": "Qwen3 е ново поколение модел на Tongyi Qianwen с значително подобрени способности, достигащи водещо ниво в индустрията в разсъждения, общи, агенти и многоезични основни способности, и поддържа превключване на режим на мислене."
   },
   "Qwen/Qwen3-8B": {
     "description": "Qwen3 е ново поколение модел на Tongyi Qianwen с значително подобрени способности, достигащи водещо ниво в индустрията в разсъждения, общи, агенти и многоезични основни способности, и поддържа превключване на режим на мислене."
   },
+  "Qwen/Qwen3-Coder-30B-A3B-Instruct": {
+    "description": "Qwen3-Coder-30B-A3B-Instruct е кодов модел от серията Qwen3, разработен от екипа Tongyi Qianwen на Alibaba. Като рационализиран и оптимизиран модел, той запазва висока производителност и ефективност, като същевременно е фокусиран върху подобряване на възможностите за обработка на код. Моделът показва значително предимство сред отворените модели при сложни задачи като агентно програмиране (Agentic Coding), автоматизирани браузърни операции и извикване на инструменти. Той предлага нативна поддръжка за дълги контексти до 256K токена и може да се разшири до 1M токена, което позволява по-добро разбиране и обработка на ниво кодова база. Освен това моделът предоставя силна поддръжка за агентно кодиране в платформи като Qwen Code и CLINE и е проектирал специален формат за извикване на функции."
+  },
+  "Qwen/Qwen3-Coder-480B-A35B-Instruct": {
+    "description": "Qwen3-Coder-480B-A35B-Instruct е публикуван от Alibaba и до момента е един от най-агентно ориентираните (agentic) кодови модели. Това е смесен експертен (MoE) модел с общо 480 милиарда параметри и 35 милиарда активни параметри, който постига баланс между ефективност и производителност. Моделът поддържа родно контекстна дължина от 256K (прибл. 260 000) токена и може да бъде екстраполиран чрез методи като YaRN до 1 милион токена, което му позволява да обработва големи кодови бази и сложни програмистки задачи. Qwen3-Coder е специално проектиран за агентно ориентирани (agentic) кодови работни потоци — той не само генерира код, но може и автономно да взаимодейства с инструменти и среди за разработка, за да решава сложни програмистки проблеми. В множество бенчмаркове за кодиране и агентни задачи моделът постига водещи резултати сред отворените модели и неговата производителност е сравнима с тази на водещи модели като Claude Sonnet 4."
+  },
   "Qwen2-72B-Instruct": {
     "description": "Qwen2 е най-новата серия на модела Qwen, поддържаща 128k контекст. В сравнение с текущите най-добри отворени модели, Qwen2-72B значително надминава водещите модели в области като разбиране на естествен език, знания, код, математика и многоезичност."
   },
@@ -368,6 +398,9 @@
   "Qwen3-235B": {
     "description": "Qwen3-235B-A22B е MoE (хибриден експертен модел), който въвежда „хибриден режим на разсъждение“, позволяващ на потребителите безпроблемно превключване между „режим мислене“ и „режим без мислене“. Поддържа разбиране и разсъждение на 119 езика и диалекта и разполага с мощни възможности за извикване на инструменти. В множество базови тестове за общи способности, кодиране, математика, многоезичност, знания и разсъждение, той може да се конкурира с водещите големи модели на пазара като DeepSeek R1, OpenAI o1, o3-mini, Grok 3 и Google Gemini 2.5 Pro."
   },
+  "Qwen3-235B-A22B-Instruct-2507-FP8": {
+    "description": "Qwen3 235B A22B Instruct 2507: Модел, оптимизиран за напреднало разсъждение и диалогови инструкции, с хибридна експертна архитектура, поддържаща ефективност при голям брой параметри."
+  },
   "Qwen3-32B": {
     "description": "Qwen3-32B е плътен модел (Dense Model), който въвежда „хибриден режим на разсъждение“, позволяващ на потребителите безпроблемно превключване между „режим мислене“ и „режим без мислене“. Благодарение на подобрения в архитектурата на модела, увеличени тренировъчни данни и по-ефективни методи за обучение, общата производителност е сравнима с тази на Qwen2.5-72B."
   },
@@ -824,6 +857,9 @@
   "deepseek-ai/DeepSeek-V3": {
     "description": "DeepSeek-V3 е езиков модел с 6710 милиарда параметри, базиран на смесени експерти (MoE), който използва многоглаво потенциално внимание (MLA) и архитектурата DeepSeekMoE, комбинирайки стратегии за баланс на натоварването без помощни загуби, за да оптимизира производителността на извеждане и обучение. Чрез предварително обучение на 14.8 трилиона висококачествени токени и последващо наблюдавано фино настройване и подсилено обучение, DeepSeek-V3 надминава производителността на други отворени модели и се приближава до водещите затворени модели."
   },
+  "deepseek-ai/DeepSeek-V3.1": {
+    "description": "DeepSeek-V3.1 е хибриден голям езиков модел, пуснат от DeepSeek AI, който включва множество важни подобрения спрямо предишните версии. Основната иновация на модела е интеграцията на „режим на мислене“ (Thinking Mode) и „режим без мислене“ (Non-thinking Mode), които потребителите могат гъвкаво да превключват чрез настройка на чат шаблони, за да отговарят на различни задачи. След специална пост-тренировка, V3.1 значително подобрява производителността при използване на инструменти и задачи на агенти, като по-добре поддържа външни търсачки и изпълнение на сложни многостъпкови задачи. Моделът е дообучен върху DeepSeek-V3.1-Base чрез двуфазен метод за разширяване на дълги текстове, което значително увеличава обема на тренировъчните данни и подобрява работата с дълги документи и кодове. Като отворен модел, DeepSeek-V3.1 демонстрира способности, сравними с водещи затворени модели в области като кодиране, математика и разсъждение, като същевременно с хибридната си експертна (MoE) архитектура поддържа голям капацитет на модела и ефективно намалява разходите за изчисления."
+  },
   "deepseek-ai/deepseek-llm-67b-chat": {
     "description": "DeepSeek 67B е напреднал модел, обучен за диалози с висока сложност."
   },
@@ -894,7 +930,7 @@
     "description": "DeepSeek R1 пълна версия, с 671B параметри, поддържаща търсене в реално време, с по-силни способности за разбиране и генериране."
   },
   "deepseek-reasoner": {
-    "description": "Модел за извеждане, разработен от DeepSeek. Преди да предостави окончателния отговор, моделът първо извежда част от веригата на мислене, за да повиши точността на крайния отговор."
+    "description": "DeepSeek V3.1 режим на мислене. Преди да изведе окончателния отговор, моделът първо генерира мисловна верига, за да повиши точността на крайния отговор."
   },
   "deepseek-v2": {
     "description": "DeepSeek V2 е ефективен модел на Mixture-of-Experts, подходящ за икономически ефективни нужди от обработка."
@@ -908,12 +944,18 @@
   "deepseek-v3-0324": {
     "description": "DeepSeek-V3-0324 е MoE модел с 671B параметри, който се отличава с предимства в програмирането и техническите способности, разбирането на контекста и обработката на дълги текстове."
   },
+  "deepseek-v3.1": {
+    "description": "DeepSeek-V3.1 е новият хибриден модел за разсъждение на DeepSeek, който поддържа два режима на разсъждение: мислене и немислене, с по-висока ефективност на мислене в сравнение с DeepSeek-R1-0528. След оптимизация чрез пост-тренировка, използването на агентски инструменти и изпълнението на задачи от интелигентни агенти са значително подобрени. Поддържа контекстен прозорец до 128k и максимална дължина на изхода до 64k токена."
+  },
   "deepseek/deepseek-chat-v3-0324": {
     "description": "DeepSeek V3 е експертен смесен модел с 685B параметри, последната итерация на флагманската серия чат модели на екипа DeepSeek.\n\nТой наследява модела [DeepSeek V3](/deepseek/deepseek-chat-v3) и показва отлични резултати в различни задачи."
   },
   "deepseek/deepseek-chat-v3-0324:free": {
     "description": "DeepSeek V3 е експертен смесен модел с 685B параметри, последната итерация на флагманската серия чат модели на екипа DeepSeek.\n\nТой наследява модела [DeepSeek V3](/deepseek/deepseek-chat-v3) и показва отлични резултати в различни задачи."
   },
+  "deepseek/deepseek-chat-v3.1": {
+    "description": "DeepSeek-V3.1 е голям хибриден модел за разсъждение, който поддържа 128K дълъг контекст и ефективно превключване на режими, постигащ изключителна производителност и скорост при използване на инструменти, генериране на код и сложни задачи за разсъждение."
+  },
   "deepseek/deepseek-r1": {
     "description": "DeepSeek-R1 значително подобри способността на модела за разсъждение при наличието на много малко маркирани данни. Преди да предостави окончателния отговор, моделът първо ще изведе част от съдържанието на веригата на мислене, за да повиши точността на окончателния отговор."
   },
@@ -1103,12 +1145,27 @@
   "flux-dev": {
     "description": "FLUX.1 [dev] е отворен и пречистен модел, предназначен за нетърговска употреба. Той запазва качество на изображенията и способността за следване на инструкции, близки до професионалната версия на FLUX, като същевременно предлага по-висока ефективност на работа и по-добро използване на ресурсите в сравнение със стандартни модели със същия размер."
   },
+  "flux-kontext-max": {
+    "description": "Най-съвременно генериране и редактиране на контекстуални изображения — комбиниране на текст и изображения за постигане на прецизни и кохерентни резултати."
+  },
+  "flux-kontext-pro": {
+    "description": "Най-съвременни възможности за контекстно генериране и редактиране на изображения — комбиниране на текст и изображения за постигане на прецизни и последователни резултати."
+  },
   "flux-kontext/dev": {
     "description": "FLUX.1 модел, фокусиран върху задачи за редактиране на изображения, поддържащ текстови и визуални входни данни."
   },
   "flux-merged": {
     "description": "FLUX.1-merged комбинира дълбоките характеристики, изследвани в разработката на \"DEV\" версията, с високоскоростните предимства на \"Schnell\". Тази комбинация не само разширява границите на производителността на модела, но и увеличава обхвата на неговото приложение."
   },
+  "flux-pro": {
+    "description": "Водещ комерсиален модел за генериране на изображения с изкуствен интелект — несравнимо качество на изображенията и богато разнообразие на генерираните резултати."
+  },
+  "flux-pro-1.1": {
+    "description": "Усъвършенстван професионален модел за генериране на изображения с изкуствен интелект — предлага изключително качество на изображенията и прецизно изпълнение на подадените подсказки."
+  },
+  "flux-pro-1.1-ultra": {
+    "description": "Генериране на изображения с изкуствен интелект с изключително висока резолюция — поддържа изход 4 мегапиксела, създава ултраясни изображения за по-малко от 10 секунди."
+  },
   "flux-pro/kontext": {
     "description": "FLUX.1 Kontext [pro] може да обработва текст и референтни изображения като вход, осигурявайки безпроблемно целенасочено локално редактиране и сложни трансформации на цялостната сцена."
   },
@@ -1193,6 +1250,12 @@
   "gemini-2.5-flash": {
     "description": "Gemini 2.5 Flash е най-ефективният модел на Google, предлагащ пълна функционалност."
   },
+  "gemini-2.5-flash-image-preview": {
+    "description": "Nano Banana е най-новият, най-бързият и най-ефективният роден мултимодален модел на Google, който ви позволява да генерирате и редактирате изображения чрез диалог."
+  },
+  "gemini-2.5-flash-image-preview:image": {
+    "description": "Nano Banana е най-новият, най-бързият и най-ефективният роден мултимодален модел на Google, който ви позволява да генерирате и редактирате изображения чрез диалог."
+  },
   "gemini-2.5-flash-lite": {
     "description": "Gemini 2.5 Flash-Lite е най-малкият и най-ефективен модел на Google, създаден специално за масово използване."
   },
@@ -1295,6 +1358,9 @@
   "glm-4.5-x": {
     "description": "Експресната версия на GLM-4.5, която съчетава силна производителност с генериране на скорост до 100 токена в секунда."
   },
+  "glm-4.5v": {
+    "description": "Новото поколение визуален модел за разсъждение на Zhipu, базиран на MOE архитектура, с общо 106B параметри и 12B активни параметри, постига SOTA сред отворените мултимодални модели в своя клас в различни бенчмаркове, обхващайки често срещани задачи като обработка на изображения, видео, разбиране на документи и GUI задачи."
+  },
   "glm-4v": {
     "description": "GLM-4V предлага мощни способности за разбиране и разсъждение на изображения, поддържаща множество визуални задачи."
   },
@@ -1331,6 +1397,12 @@
   "google/gemini-2.5-flash": {
     "description": "Gemini 2.5 Flash е най-усъвършенстваният основен модел на Google, специално проектиран за напреднали задачи по разсъждение, кодиране, математика и наука. Той включва вградена способност за „мислене“, която му позволява да предоставя отговори с по-висока точност и по-детайлна обработка на контекста.\n\nЗабележка: Този модел има два варианта: с мислене и без мислене. Ценообразуването на изхода се различава значително в зависимост от това дали способността за мислене е активирана. Ако изберете стандартния вариант (без суфикса „:thinking“), моделът ясно избягва генерирането на мисловни токени.\n\nЗа да използвате способността за мислене и да получавате мисловни токени, трябва да изберете варианта „:thinking“, което ще доведе до по-висока цена за изход с мислене.\n\nОсвен това, Gemini 2.5 Flash може да бъде конфигуриран чрез параметъра „максимален брой токени за разсъждение“, както е описано в документацията (https://openrouter.ai/docs/use-cases/reasoning-tokens#max-tokens-for-reasoning)."
   },
+  "google/gemini-2.5-flash-image-preview": {
+    "description": "Gemini 2.5 Flash експериментален модел, поддържащ генериране на изображения."
+  },
+  "google/gemini-2.5-flash-image-preview:free": {
+    "description": "Gemini 2.5 Flash експериментален модел, поддържащ генериране на изображения."
+  },
   "google/gemini-2.5-flash-preview": {
     "description": "Gemini 2.5 Flash е най-напредналият основен модел на Google, проектиран за напреднали разсъждения, кодиране, математика и научни задачи. Той включва вградена способност за \"мислене\", което му позволява да предоставя отговори с по-висока точност и детайлна обработка на контекста.\n\nЗабележка: Този модел има два варианта: с мислене и без мислене. Цените на изхода значително варират в зависимост от активирането на способността за мислене. Ако изберете стандартния вариант (без суфикс \":thinking\"), моделът ще избягва генерирането на токени за мислене.\n\nЗа да се възползвате от способността за мислене и да получите токени за мислене, трябва да изберете варианта \":thinking\", което ще доведе до по-високи цени на изхода за мислене.\n\nОсвен това, Gemini 2.5 Flash може да бъде конфигуриран чрез параметъра \"максимален брой токени за разсъждение\", както е описано в документацията (https://openrouter.ai/docs/use-cases/reasoning-tokens#max-tokens-for-reasoning)."
   },
@@ -1434,7 +1506,7 @@
     "description": "GPT-4.1 mini предлага баланс между интелигентност, скорост и разходи, което го прави привлекателен модел за много случаи на употреба."
   },
   "gpt-4.5-preview": {
-    "description": "Изследователската предварителна версия на GPT-4.5, която е нашият най-голям и мощен GPT модел до момента. Тя притежава обширни знания за света и може по-добре да разбира намеренията на потребителите, което я прави изключително ефективна в креативни задачи и автономно планиране. GPT-4.5 приема текстови и изображен вход и генерира текстови изход (включително структурирани изходи). Поддържа ключови функции за разработчици, като извикване на функции, пакетно API и потоков изход. В задачи, изискващи креативно, открито мислене и диалог (като писане, учене или изследване на нови идеи), GPT-4.5 показва особени способности. Крайната дата на знанията е октомври 2023."
+    "description": "GPT-4.5-preview е най-новият универсален модел, който притежава задълбочени световни познания и по-добро разбиране на намеренията на потребителите, отличава се в творчески задачи и при планирането на агенти. Знанията на модела са актуални до октомври 2023 г."
   },
   "gpt-4o": {
     "description": "ChatGPT-4o е динамичен модел, който се актуализира в реално време, за да поддържа най-новата версия. Той комбинира мощно разбиране на езика и генериране на текст, подходящ за мащабни приложения, включително обслужване на клиенти, образование и техническа поддръжка."
@@ -1502,6 +1574,9 @@
   "gpt-oss": {
     "description": "GPT-OSS 20B е отворен голям езиков модел, публикуван от OpenAI, използващ технологията за квантуване MXFP4, подходящ за работа на висок клас потребителски GPU или Apple Silicon Mac. Този модел се отличава с отлични резултати в генерирането на диалози, писането на код и задачи за разсъждение, като поддържа извикване на функции и използване на инструменти."
   },
+  "gpt-oss-120b": {
+    "description": "GPT-OSS-120B MXFP4 квантизиран трансформър модел, който запазва силна производителност при ограничени ресурси."
+  },
   "gpt-oss:120b": {
     "description": "GPT-OSS 120B е голям отворен езиков модел, публикуван от OpenAI, използващ технологията за квантуване MXFP4, предназначен за флагмански клас модели. Изисква многократни GPU или високопроизводителна работна станция за работа, с изключителни възможности в сложни разсъждения, генериране на код и многоезична обработка, поддържайки усъвършенствано извикване на функции и интеграция на инструменти."
   },
@@ -1529,6 +1604,9 @@
   "grok-4": {
     "description": "Нашият най-нов и най-мощен флагмански модел, който се отличава с изключителни резултати в обработката на естествен език, математическите изчисления и разсъжденията — перфектен универсален играч."
   },
+  "grok-code-fast-1": {
+    "description": "С удоволствие представяме grok-code-fast-1, бърз и икономичен модел за извод, който се отличава с отлични резултати при кодиране на агенти."
+  },
   "gryphe/mythomax-l2-13b": {
     "description": "MythoMax l2 13B е езиков модел, който комбинира креативност и интелигентност, обединявайки множество водещи модели."
   },
@@ -1637,9 +1715,18 @@
   "image-01-live": {
     "description": "Модел за генериране на изображения с фини детайли, поддържащ генериране от текст и настройка на стил."
   },
+  "imagen-4.0-fast-generate-001": {
+    "description": "Imagen — серия модели за преобразуване от текст към изображение от 4-то поколение, бърза версия"
+  },
+  "imagen-4.0-generate-001": {
+    "description": "Imagen: серия от модели от 4-то поколение за генериране на изображения от текст"
+  },
   "imagen-4.0-generate-preview-06-06": {
     "description": "Imagen 4-то поколение текст-към-изображение модел серия"
   },
+  "imagen-4.0-ultra-generate-001": {
+    "description": "Imagen, 4-то поколение модел за преобразуване на текст в изображение, серия Ultra"
+  },
   "imagen-4.0-ultra-generate-preview-06-06": {
     "description": "Imagen 4-то поколение текст-към-изображение модел серия Ултра версия"
   },
@@ -1679,6 +1766,9 @@
   "kimi-k2-0711-preview": {
     "description": "kimi-k2 е базов модел с MoE архитектура с изключителни способности за кодиране и агентски функции, с общо 1 трилион параметри и 32 милиарда активни параметри. В тестове за общо знание, програмиране, математика и агентски задачи, моделът K2 превъзхожда други водещи отворени модели."
   },
+  "kimi-k2-turbo-preview": {
+    "description": "Kimi-k2 е базов модел с MoE архитектура, който притежава изключителни възможности за работа с код и агентни функции. Общият брой параметри е 1T, а активните параметри са 32B. В бенчмарковете за основни категории като общо знание и разсъждение, програмиране, математика и агентни задачи, моделът K2 превъзхожда другите водещи отворени модели."
+  },
   "kimi-latest": {
     "description": "Kimi интелигентен асистент използва най-новия Kimi голям модел, който може да съдържа нестабилни функции. Поддържа разбиране на изображения и автоматично избира 8k/32k/128k модел за таксуване в зависимост от дължината на контекста на заявката."
   },
@@ -1763,6 +1853,9 @@
   "llava:34b": {
     "description": "LLaVA е многомодален модел, комбиниращ визуален кодер и Vicuna, предназначен за мощно визуално и езиково разбиране."
   },
+  "magistral-medium-latest": {
+    "description": "Magistral Medium 1.1 е водещ модел за инференция, публикуван от Mistral AI през юли 2025 г."
+  },
   "mathstral": {
     "description": "MathΣtral е проектиран за научни изследвания и математически разсъждения, предоставяйки ефективни изчислителни способности и интерпретация на резултати."
   },
@@ -2094,7 +2187,7 @@
     "description": "o1-mini е бърз и икономичен модел за изводи, проектиран за приложения в програмирането, математиката и науката. Моделът разполага с контекст от 128K и дата на знание до октомври 2023."
   },
   "o1-preview": {
-    "description": "o1 е новият модел за изводи на OpenAI, подходящ за сложни задачи, изискващи обширни общи знания. Моделът разполага с контекст от 128K и дата на знание до октомври 2023."
+    "description": "Фокусиран върху усъвършенствано разсъждение и решаване на сложни проблеми, включително математически и научни задачи. Отлично подходящ за приложения, които изискват дълбоко разбиране на контекста и автономни работни процеси."
   },
   "o1-pro": {
     "description": "Моделите от серията o1 са обучени чрез подсилващо обучение, което им позволява да мислят преди да отговорят и да изпълняват сложни задачи за разсъждение. Моделът o1-pro използва повече изчислителни ресурси за по-задълбочено мислене, осигурявайки постоянно по-високо качество на отговорите."
@@ -2213,8 +2306,14 @@
   "qwen-coder-turbo-latest": {
     "description": "Моделът на кода Qwen."
   },
+  "qwen-flash": {
+    "description": "Серията \"Tongyi Qianwen\" включва най-бързите и с изключително ниски разходи модели, подходящи за прости задачи."
+  },
   "qwen-image": {
-    "description": "Мощен модел за генериране на изображения от екипа на Qwen, с впечатляващи възможности за генериране на китайски текст и разнообразни визуални стилове на изображения."
+    "description": "Qwen-Image е универсален модел за генериране на изображения, който поддържа множество художествени стилове и е особено добър в рендериране на сложни текстове, включително на китайски и английски. Моделът поддържа многоредови оформления, генериране на текст на ниво абзац и изобразяване на детайли с висока прецизност, позволявайки създаване на сложни комбинирани оформления от изображение и текст."
+  },
+  "qwen-image-edit": {
+    "description": "Професионален модел за редактиране на изображения, публикуван от екипа на Qwen, който поддържа семантично редактиране и редактиране на външния вид и може прецизно да обработва текст на китайски и английски, извършвайки висококачествени редакции на изображения като трансформация на стил и въртене на обекти."
   },
   "qwen-long": {
     "description": "Qwen е мащабен езиков модел, който поддържа дълги текстови контексти и диалогови функции, базирани на дълги документи и множество документи."
@@ -2241,7 +2340,7 @@
     "description": "通义千问（Qwen） е подобрена версия на мащабен езиков модел, който поддържа вход на различни езици, включително китайски и английски."
   },
   "qwen-turbo": {
-    "description": "通义千问（Qwen） е мащабен езиков модел, който поддържа вход на различни езици, включително китайски и английски."
+    "description": "Версията „Tongyi Qianwen Turbo“ няма да получава повече актуализации; препоръчваме да я замените с „Tongyi Qianwen Flash“. Tongyi Qianwen е много голям езиков модел, който поддържа въвеждане на китайски, английски и други езици."
   },
   "qwen-vl-chat-v1": {
     "description": "Qwen VL поддържа гъвкави интерактивни методи, включително множество изображения, многократни въпроси и отговори, творчество и др."
@@ -2558,9 +2657,15 @@
   "step-2x-large": {
     "description": "Новото поколение модел за генериране на изображения Step Star, специализиран в генериране на висококачествени изображения според текстови описания от потребителя. Новият модел създава по-реалистични текстури и има по-силни способности за генериране на китайски и английски текст."
   },
+  "step-3": {
+    "description": "Този модел притежава мощни способности за визуално възприятие и сложни разсъждения. Той може точно да извършва междудисциплинарно разбиране на сложни знания, съвместен анализ на математическа и визуална информация, както и да решава различни визуални аналитични задачи от ежедневието."
+  },
   "step-r1-v-mini": {
     "description": "Този модел е мощен модел за разсъждение с отлични способности за разбиране на изображения, способен да обработва информация от изображения и текст, и след дълбочинно разсъждение да генерира текстово съдържание. Моделът показва изключителни резултати в областта на визуалните разсъждения, като същевременно притежава първокласни способности в математиката, кода и текстовите разсъждения. Дължината на контекста е 100k."
   },
+  "stepfun-ai/step3": {
+    "description": "Step3 е авангарден мултимодален модел за разсъждение, публикуван от StepFun (阶跃星辰). Той е изграден върху архитектура на смес от експерти (MoE) с общо 321 милиарда параметъра и 38 милиарда активни параметъра. Моделът е с енд-ту-енд дизайн, целящ минимизиране на разходите за декодиране, като същевременно предоставя водещи резултати във визуално-лингвистичното разсъждение. Чрез кооперативния дизайн на многоматрично факторизирано внимание (MFA) и декуплиране на внимание и FFN (AFD), Step3 поддържа отлична ефективност както на флагмански, така и на по-бюджетни ускорители. По време на предварителното обучение Step3 е обработил над 20 трилиона текстови токена и 4 трилиона смесени текстово-изображенчески токена, обхващайки повече от десет езика. Моделът постига водещи резултати сред отворените модели в множество бенчмаркове, включително математика, код и мултимодални задачи."
+  },
   "taichu_llm": {
     "description": "Моделът на езика TaiChu е с изключителни способности за разбиране на езика, текстово генериране, отговори на знания, програмиране, математически изчисления, логическо разсъждение, анализ на емоции, резюмиране на текст и др. Иновативно комбинира предварително обучение с големи данни и разнообразни източници на знания, чрез непрекъснато усъвършенстване на алгоритмичните технологии и усвояване на нови знания от масивни текстови данни, за да осигури на потребителите по-удобна информация и услуги, както и по-интелигентно изживяване."
   },
@@ -2588,18 +2693,9 @@
   "thudm/glm-4-9b-chat": {
     "description": "GLM-4 е последната версия на предварително обучен модел от серията, публикувана от Zhizhu AI."
   },
-  "thudm/glm-4-9b:free": {
-    "description": "GLM-4-9B-0414 е езиков модел с 9 милиарда параметри от серията GLM-4, разработен от THUDM. GLM-4-9B-0414 използва същите стратегии за усилено обучение и подравняване, които се прилагат за по-голямата му 32B версия, за да постигне висока производителност в съотношение с размера си, което го прави подходящ за внедряване с ограничени ресурси, което все пак изисква силни способности за разбиране и генериране на език."
-  },
   "thudm/glm-z1-32b": {
     "description": "GLM-Z1-32B-0414 е подобрена версия на GLM-4-32B, проектирана за дълбока математика, логика и решаване на проблеми, свързани с код. Той прилага разширено обучение с подсилване (за специфични задачи и на базата на общи предпочитания) за подобряване на производителността при сложни многостепенни задачи. В сравнение с основния модел GLM-4-32B, Z1 значително подобрява способностите в структурираното разсъждение и формалните области.\n\nМоделът поддържа прилагане на стъпки за \"мислене\" чрез инженеринг на подсказки и предлага подобрена последователност за дълги формати на изхода. Той е оптимизиран за работни потоци на агенти и поддържа дълъг контекст (чрез YaRN), извиквания на JSON инструменти и конфигурации за фино проби за стабилно разсъждение. Идеален е за случаи, изискващи дълбочинно разсъждение, многостепенни разсъждения или формализирани изводи."
   },
-  "thudm/glm-z1-32b:free": {
-    "description": "GLM-Z1-32B-0414 е подобрена версия на GLM-4-32B, проектирана за дълбока математика, логика и решаване на проблеми, свързани с код. Той прилага разширено обучение с подсилване (за специфични задачи и на базата на общи предпочитания) за подобряване на производителността при сложни многостепенни задачи. В сравнение с основния модел GLM-4-32B, Z1 значително подобрява способностите в структурираното разсъждение и формалните области.\n\nМоделът поддържа прилагане на стъпки за \"мислене\" чрез инженеринг на подсказки и предлага подобрена последователност за дълги формати на изхода. Той е оптимизиран за работни потоци на агенти и поддържа дълъг контекст (чрез YaRN), извиквания на JSON инструменти и конфигурации за фино проби за стабилно разсъждение. Идеален е за случаи, изискващи дълбочинно разсъждение, многостепенни разсъждения или формализирани изводи."
-  },
-  "thudm/glm-z1-9b:free": {
-    "description": "GLM-Z1-9B-0414 е езиков модел с 9B параметри от серията GLM-4, разработен от THUDM. Той прилага технологии, първоначално използвани в по-големия GLM-Z1 модел, включително разширено усилено обучение, подравняване на двойки и обучение за интензивни разсъждения в области като математика, кодиране и логика. Въпреки по-малкия си размер, той показва силна производителност в общите задачи за разсъждение и надминава много от отворените модели на нивото на теглата."
-  },
   "thudm/glm-z1-rumination-32b": {
     "description": "THUDM: GLM Z1 Rumination 32B е дълбок разсъдъчен модел с 32B параметри от серията GLM-4-Z1, оптимизиран за сложни, отворени задачи, изискващи дълго разсъждение. Той е построен на основата на glm-4-32b-0414, с добавени допълнителни етапи на усилено обучение и многостепенни стратегии за подравняване, въвеждайки \"разсъждателни\" способности, предназначени да симулират разширена когнитивна обработка. Това включва итеративно разсъждение, многократен анализ и работни потоци, подобрени с инструменти, като търсене, извличане и синтез с осведоменост за цитати.\n\nМоделът показва отлични резултати в изследователското писане, сравнителния анализ и сложните въпроси и отговори. Той поддържа извиквания на функции за търсене и навигация (\"search\", \"click\", \"open\", \"finish\"), което позволява използването му в агенти. Разсъждателното поведение се контролира от многократни цикли с базирани на правила награди и механизми за забавено вземане на решения, с референтни рамки за дълбоки изследвания, като вътрешния стек за подравняване на OpenAI. Този вариант е подходящ за сценарии, изискващи дълбочина, а не скорост."
   },
@@ -2707,5 +2803,8 @@
   },
   "zai-org/GLM-4.5-Air": {
     "description": "GLM-4.5-Air е базов модел, специално създаден за интелигентни агенти, използващ архитектура с микс от експерти (Mixture-of-Experts). Той е дълбоко оптимизиран за използване на инструменти, уеб браузване, софтуерно инженерство и фронтенд програмиране, и поддържа безпроблемна интеграция с кодови агенти като Claude Code и Roo Code. GLM-4.5 използва смесен режим на разсъждение, подходящ за сложни и ежедневни приложения."
+  },
+  "zai-org/GLM-4.5V": {
+    "description": "GLM-4.5V е най-новото поколение визуално-езиков модел (VLM), публикуван от Zhipu AI (智谱 AI). Моделът е изграден върху водещия текстов модел GLM-4.5-Air, който разполага с общо 106 милиарда параметри и 12 милиарда активационни параметри, и използва архитектура с разбъркани експерти (Mixture of Experts, MoE), целяща постигане на висока производителност при по-ниски разходи за инференция. Технически GLM-4.5V продължава линията на GLM-4.1V-Thinking и въвежда иновации като триизмерно ротационно позиционно кодиране (3D-RoPE), което значително засилва възприемането и разсъжденията относно триизмерните пространствени взаимовръзки. Чрез оптимизации в етапите на предварително обучение, супервизирано фино настройване и подсилено обучение, моделът може да обработва различни визуални формати — изображения, видео и дълги документи — и в 41 публични мултимодални бенчмарка достига водещи резултати сред отворените модели от същия клас. Освен това моделът добавя превключвател за 'режим на мислене', който позволява на потребителите гъвкаво да избират между бърз отговор и дълбоко разсъждение, за да балансират ефективността и качеството."
   }
 }

locales/bg-BG/providers.json

@@ -11,6 +11,9 @@
   "aihubmix": {
     "description": "AiHubMix предоставя достъп до множество AI модели чрез единен API интерфейс."
   },
+  "akashchat": {
+    "description": "Akash е пазар за облачни ресурси без нужда от разрешение, който предлага конкурентни цени в сравнение с традиционните облачни доставчици."
+  },
   "anthropic": {
     "description": "Anthropic е компания, специализирана в изследвания и разработка на изкуствен интелект, предлагаща набор от напреднали езикови модели, като Claude 3.5 Sonnet, Claude 3 Sonnet, Claude 3 Opus и Claude 3 Haiku. Тези модели постигат идеален баланс между интелигентност, скорост и разходи, подходящи за различни приложения, от корпоративни натоварвания до бързи отговори. Claude 3.5 Sonnet, като най-новия им модел, показва отлични резултати в множество оценки, като същевременно поддържа висока цена-качество."
   },
@@ -26,6 +29,9 @@
   "bedrock": {
     "description": "Bedrock е услуга, предоставяна от Amazon AWS, фокусирана върху предоставянето на напреднали AI езикови и визуални модели за предприятия. Семейството на моделите включва серията Claude на Anthropic, серията Llama 3.1 на Meta и други, обхващащи разнообразие от опции от леки до високо производителни, поддържащи текстово генериране, диалог, обработка на изображения и много други задачи, подходящи за различни мащаби и нужди на бизнес приложения."
   },
+  "bfl": {
+    "description": "Водеща изследователска лаборатория за авангарден изкуствен интелект, която изгражда визуалната инфраструктура на утрешния ден."
+  },
   "cloudflare": {
     "description": "Работа с модели на машинно обучение, задвижвани от безсървърни GPU, в глобалната мрежа на Cloudflare."
   },
@@ -86,6 +92,12 @@
   "moonshot": {
     "description": "Moonshot е отворена платформа, представена от Beijing Dark Side Technology Co., Ltd., предлагаща множество модели за обработка на естествен език, с широко приложение, включително, но не само, създаване на съдържание, академични изследвания, интелигентни препоръки, медицинска диагностика и др., поддържаща обработка на дълги текстове и сложни генериращи задачи."
   },
+  "nebius": {
+    "description": "Nebius предоставя високопроизводителна инфраструктура за глобалните иноватори в областта на изкуствения интелект чрез изграждане на мащабни GPU клъстери и вертикално интегрирана облачна платформа."
+  },
+  "newapi": {
+    "description": "Отворена платформа за обединено препращане на множество AI услуги"
+  },
   "novita": {
     "description": "Novita AI е платформа, предлагаща API услуги за множество големи езикови модели и генериране на AI изображения, гъвкава, надеждна и икономически ефективна. Поддържа най-новите отворени модели, като Llama3 и Mistral, и предлага цялостни, потребителски приятелски и автоматично разширяеми API решения за разработка на генеративни AI приложения, подходящи за бързото развитие на AI стартъпи."
   },

locales/bg-BG/setting.json

@@ -237,6 +237,10 @@
       "title": "Ограничаване на броя на съобщенията в историята",
       "unlimited": "Неограничен брой съобщения в историята"
     },
+    "enableStreaming": {
+      "desc": "Активирайте поточно предаване, за да показвате отговорите в реално време. Ако е деактивирано, ще се показва само пълният отговор.",
+      "title": "Активиране на поточно предаване"
+    },
     "historyCount": {
       "desc": "Брой исторически съобщения, носени с всяка заявка",
       "title": "Брой прикачени съобщения в историята"

locales/de-DE/chat.json

@@ -52,7 +52,11 @@
     "thinking": {
       "title": "Tiefdenk-Schalter"
     },
-    "title": "Modell Erweiterungsfunktionen"
+    "title": "Modell Erweiterungsfunktionen",
+    "urlContext": {
+      "desc": "Wenn aktiviert, werden Webseiten-Links automatisch analysiert, um den tatsächlichen Webseiteninhalt zu erfassen",
+      "title": "Webseiten-Link-Inhalt extrahieren"
+    }
   },
   "history": {
     "title": "Der Assistent wird nur die letzten {{count}} Nachrichten speichern"
@@ -125,6 +129,7 @@
       "inputWriteCached": "Eingabe Cache schreiben",
       "output": "Ausgabe",
       "outputAudio": "Audioausgabe",
+      "outputImage": "Bildausgabe",
       "outputText": "Text-Ausgabe",
       "outputTitle": "Ausgabedetails",
       "reasoning": "Tiefes Denken",
@@ -268,7 +273,8 @@
     "clientMode": {
       "actionFiletip": "Datei hochladen",
       "actionTooltip": "Hochladen",
-      "disabled": "Das aktuelle Modell unterstützt keine visuelle Erkennung und Dateianalyse. Bitte wechseln Sie das Modell, um diese Funktionen zu nutzen."
+      "disabled": "Das aktuelle Modell unterstützt keine visuelle Erkennung und Dateianalyse. Bitte wechseln Sie das Modell, um diese Funktionen zu nutzen.",
+      "visionNotSupported": "Das aktuelle Modell unterstützt keine visuelle Erkennung. Bitte wechseln Sie das Modell, um diese Funktion zu nutzen."
     },
     "preview": {
       "prepareTasks": "Vorbereitung der Teile...",

locales/de-DE/common.json

@@ -32,6 +32,7 @@
   "blog": "Produkt-Blog",
   "branching": "Unterthema erstellen",
   "branchingDisable": "Die Funktion „Unterthema“ ist nur in der Serverversion verfügbar. Wenn Sie diese Funktion benötigen, wechseln Sie bitte in den Serverbereitstellungsmodus oder verwenden Sie LobeChat Cloud.",
+  "branchingRequiresSavedTopic": "Das aktuelle Thema wurde nicht gespeichert. Speichern Sie es, um die Unterthemenfunktion nutzen zu können.",
   "cancel": "Abbrechen",
   "changelog": "Änderungsprotokoll",
   "clientDB": {
@@ -182,6 +183,13 @@
     "title": "Mögen Sie unser Produkt?"
   },
   "fullscreen": "Vollbildmodus",
+  "geminiImageChineseWarning": {
+    "content": "Nano Banana kann bei der Verwendung von Chinesisch möglicherweise keine Bilder generieren. Es wird empfohlen, Englisch zu verwenden, um bessere Ergebnisse zu erzielen.",
+    "continueGenerate": "Weiter generieren",
+    "continueSend": "Weiter senden",
+    "doNotShowAgain": "Nicht mehr anzeigen",
+    "title": "Hinweis zur chinesischen Eingabe"
+  },
   "historyRange": "Verlaufsbereich",
   "import": "Importieren",
   "importData": "Daten importieren",

locales/de-DE/components.json

@@ -1,4 +1,9 @@
 {
+  "ArgsInput": {
+    "addArgument": "Parameter hinzufügen",
+    "argumentPlaceholder": "Parameter {{index}}",
+    "enterFirstArgument": "Geben Sie den ersten Parameter ein..."
+  },
   "DragUpload": {
     "dragDesc": "Ziehen Sie Dateien hierher, um mehrere Bilder hochzuladen.",
     "dragFileDesc": "Ziehen Sie Bilder und Dateien hierher, um mehrere Bilder und Dateien hochzuladen.",
@@ -68,6 +73,18 @@
   "GoBack": {
     "back": "Zurück"
   },
+  "HtmlPreview": {
+    "actions": {
+      "download": "Herunterladen",
+      "preview": "Vorschau"
+    },
+    "iframeTitle": "HTML-Vorschau",
+    "mode": {
+      "code": "Code",
+      "preview": "Vorschau"
+    },
+    "title": "HTML-Vorschau"
+  },
   "ImageUpload": {
     "actions": {
       "changeImage": "Klicken, um das Bild zu ändern",
@@ -125,6 +142,12 @@
     },
     "progress": {
       "uploadingWithCount": "{{completed}}/{{total}} hochgeladen"
+    },
+    "validation": {
+      "fileSizeExceeded": "Maximale Dateigröße überschritten",
+      "fileSizeExceededDetail": "{{fileName}} ({{actualSize}}) überschreitet die maximal zulässige Größe von {{maxSize}}",
+      "fileSizeExceededMultiple": "{{count}} Dateien überschreiten die maximal zulässige Größe von {{maxSize}}: {{fileList}}",
+      "imageCountExceeded": "Maximale Anzahl an Bildern überschritten"
     }
   },
   "OllamaSetupGuide": {

locales/de-DE/error.json

@@ -85,6 +85,17 @@
     "CreateMessageError": "Es tut uns leid, die Nachricht konnte nicht erfolgreich gesendet werden. Bitte kopieren Sie den Inhalt und senden Sie ihn erneut. Nach dem Aktualisieren der Seite wird diese Nachricht nicht gespeichert.",
     "ExceededContextWindow": "Der aktuelle Anfrageinhalt überschreitet die von dem Modell verarbeitbare Länge. Bitte reduzieren Sie die Menge des Inhalts und versuchen Sie es erneut.",
     "FreePlanLimit": "Sie sind derzeit ein kostenloser Benutzer und können diese Funktion nicht nutzen. Bitte aktualisieren Sie auf ein kostenpflichtiges Abonnement, um fortzufahren.",
+    "GoogleAIBlockReason": {
+      "BLOCKLIST": "Ihr Inhalt enthält verbotene Wörter. Bitte prüfen und ändern Sie Ihre Eingabe und versuchen Sie es erneut.",
+      "IMAGE_SAFETY": "Die Bildgenerierung wurde aus Sicherheitsgründen blockiert. Bitte passen Sie Ihre Anfrage zur Bildgenerierung an und versuchen Sie es erneut.",
+      "LANGUAGE": "Die von Ihnen verwendete Sprache wird derzeit nicht unterstützt. Bitte versuchen Sie es erneut auf Englisch oder in einer anderen unterstützten Sprache.",
+      "OTHER": "Der Inhalt wurde aus unbekannten Gründen blockiert. Bitte formulieren Sie Ihre Anfrage neu.",
+      "PROHIBITED_CONTENT": "Ihre Anfrage könnte verbotene Inhalte enthalten. Bitte passen Sie Ihre Anfrage an und stellen Sie sicher, dass sie den Nutzungsrichtlinien entspricht.",
+      "RECITATION": "Ihr Inhalt wurde wegen möglicher Urheberrechtsverletzungen blockiert. Bitte verwenden Sie Originalinhalte oder formulieren Sie Ihre Anfrage um.",
+      "SAFETY": "Ihr Inhalt wurde aufgrund von Sicherheitsrichtlinien blockiert. Bitte passen Sie Ihre Anfrage an und vermeiden Sie potenziell schädliche oder unangemessene Inhalte.",
+      "SPII": "Ihr Inhalt könnte sensible personenbezogene Daten enthalten. Zum Schutz der Privatsphäre entfernen Sie bitte diese Informationen und versuchen Sie es erneut.",
+      "default": "Inhalt blockiert: {{blockReason}}. Bitte passen Sie Ihre Anfrage an und versuchen Sie es erneut."
+    },
     "InsufficientQuota": "Es tut uns leid, das Kontingent (Quota) für diesen Schlüssel ist erreicht. Bitte überprüfen Sie Ihr Kontoguthaben oder erhöhen Sie das Kontingent des Schlüssels und versuchen Sie es erneut.",
     "InvalidAccessCode": "Das Passwort ist ungültig oder leer. Bitte geben Sie das richtige Zugangspasswort ein oder fügen Sie einen benutzerdefinierten API-Schlüssel hinzu.",
     "InvalidBedrockCredentials": "Die Bedrock-Authentifizierung ist fehlgeschlagen. Bitte überprüfen Sie AccessKeyId/SecretAccessKey und versuchen Sie es erneut.",
@@ -113,18 +124,6 @@
     "PluginServerError": "Fehler bei der Serveranfrage des Plugins. Bitte überprüfen Sie die Fehlerinformationen unten in Ihrer Plugin-Beschreibungsdatei, Plugin-Konfiguration oder Serverimplementierung",
     "PluginSettingsInvalid": "Das Plugin muss korrekt konfiguriert werden, um verwendet werden zu können. Bitte überprüfen Sie Ihre Konfiguration auf Richtigkeit",
     "ProviderBizError": "Fehler bei der Anforderung des {{provider}}-Dienstes. Bitte überprüfen Sie die folgenden Informationen oder versuchen Sie es erneut.",
-
-    "GoogleAIBlockReason": {
-      "BLOCKLIST": "Ihr Inhalt enthält verbotene Begriffe. Bitte überprüfen und ändern Sie Ihre Eingabe, bevor Sie es erneut versuchen.",
-      "IMAGE_SAFETY": "Der generierte Bildinhalt wurde aus Sicherheitsgründen blockiert. Bitte versuchen Sie, Ihre Bildgenerierungsanfrage zu ändern.",
-      "LANGUAGE": "Die von Ihnen verwendete Sprache wird nicht unterstützt. Bitte versuchen Sie es mit Englisch oder anderen unterstützten Sprachen.",
-      "OTHER": "Inhalt wurde aus unbekannten Gründen blockiert. Bitte versuchen Sie, Ihre Anfrage umzuformulieren oder wenden Sie sich an den technischen Support.",
-      "PROHIBITED_CONTENT": "Ihr Inhalt könnte verbotene Inhaltstypen enthalten. Bitte passen Sie Ihre Anfrage an, um sicherzustellen, dass sie den Nutzungsrichtlinien entspricht.",
-      "RECITATION": "Ihr Inhalt wurde aufgrund möglicher Urheberrechtsprobleme blockiert. Bitte versuchen Sie, originalen Inhalt zu verwenden oder Ihre Anfrage umzuformulieren.",
-      "SAFETY": "Ihr Inhalt wurde aufgrund von Sicherheitsrichtlinien blockiert. Bitte versuchen Sie, Ihre Anfrage anzupassen, um potenziell schädliche oder unangemessene Inhalte zu vermeiden.",
-      "SPII": "Ihr Inhalt könnte sensible persönlich identifizierbare Informationen enthalten. Zum Schutz der Privatsphäre entfernen Sie bitte relevante sensible Informationen, bevor Sie es erneut versuchen.",
-      "default": "Inhalt wurde blockiert: {{blockReason}}. Bitte passen Sie Ihren Anfrageinhalt an und versuchen Sie es erneut."
-    },
     "QuotaLimitReached": "Es tut uns leid, die aktuelle Token-Nutzung oder die Anzahl der Anfragen hat das Kontingent (Quota) für diesen Schlüssel erreicht. Bitte erhöhen Sie das Kontingent für diesen Schlüssel oder versuchen Sie es später erneut.",
     "StreamChunkError": "Fehler beim Parsen des Nachrichtenchunks der Streaming-Anfrage. Bitte überprüfen Sie, ob die aktuelle API-Schnittstelle den Standards entspricht, oder wenden Sie sich an Ihren API-Anbieter.",
     "SubscriptionKeyMismatch": "Es tut uns leid, aufgrund eines vorübergehenden Systemfehlers ist das aktuelle Abonnement vorübergehend ungültig. Bitte klicken Sie auf die Schaltfläche unten, um das Abonnement wiederherzustellen, oder kontaktieren Sie uns per E-Mail für Unterstützung.",

locales/de-DE/modelProvider.json

@@ -153,6 +153,28 @@
     "searchProviders": "Anbieter suchen...",
     "sort": "Benutzerdefinierte Sortierung"
   },
+  "newapi": {
+    "apiKey": {
+      "desc": "API-Schlüssel, der von der New API-Plattform bereitgestellt wird",
+      "placeholder": "New API API-Schlüssel",
+      "required": "API-Schlüssel ist erforderlich",
+      "title": "API-Schlüssel"
+    },
+    "apiUrl": {
+      "desc": "API-Adresse des New API-Dienstes, meistens mit /v1 am Ende",
+      "title": "API-Adresse"
+    },
+    "enabled": {
+      "title": "New API aktivieren"
+    },
+    "models": {
+      "batchSelect": "Modelle in Stapeln auswählen ({{count}} Stück)",
+      "fetch": "Modellliste abrufen",
+      "selected": "Ausgewählte Modelle",
+      "title": "Verfügbare Modelle"
+    },
+    "title": "New API"
+  },
   "ollama": {
     "checker": {
       "desc": "Testen Sie, ob die Proxy-Adresse korrekt eingetragen wurde",
@@ -185,6 +207,10 @@
     }
   },
   "providerModels": {
+    "batchSelect": {
+      "selected": "{{count}} Modelle ausgewählt",
+      "title": "Stapel-Auswahl"
+    },
     "config": {
       "aesGcm": "Ihr Schlüssel und die Proxy-Adresse werden mit dem <1>AES-GCM</1>-Verschlüsselungsalgorithmus verschlüsselt",
       "apiKey": {

locales/de-DE/models.json

@@ -74,9 +74,15 @@
   "DeepSeek-V3": {
     "description": "DeepSeek-V3 ist ein von der DeepSeek Company entwickeltes MoE-Modell. Die Ergebnisse von DeepSeek-V3 übertreffen die anderer Open-Source-Modelle wie Qwen2.5-72B und Llama-3.1-405B und stehen in der Leistung auf Augenhöhe mit den weltweit führenden Closed-Source-Modellen GPT-4o und Claude-3.5-Sonnet."
   },
+  "DeepSeek-V3-1": {
+    "description": "DeepSeek V3.1: Das nächste Generation Inferenzmodell, das komplexe Schlussfolgerungen und vernetztes Denken verbessert und sich für Aufgaben eignet, die eine tiefgehende Analyse erfordern."
+  },
   "DeepSeek-V3-Fast": {
     "description": "Modellanbieter: sophnet-Plattform. DeepSeek V3 Fast ist die Hochgeschwindigkeitsversion mit hohem TPS des DeepSeek V3 0324 Modells, voll funktionsfähig ohne Quantisierung, mit stärkerer Code- und mathematischer Leistungsfähigkeit und schnellerer Reaktionszeit!"
   },
+  "DeepSeek-V3.1": {
+    "description": "DeepSeek-V3.1 ist ein neu eingeführtes hybrides Inferenzmodell von DeepSeek, das zwei Inferenzmodi unterstützt: Denkmodus und Nicht-Denkmodus. Es ist effizienter im Denkprozess als DeepSeek-R1-0528. Durch Post-Training-Optimierung wurden die Nutzung von Agenten-Tools und die Leistung bei Agentenaufgaben erheblich verbessert."
+  },
   "Doubao-lite-128k": {
     "description": "Doubao-lite bietet extrem schnelle Reaktionszeiten und ein hervorragendes Preis-Leistungs-Verhältnis, um Kunden in verschiedenen Szenarien flexiblere Optionen zu bieten. Unterstützt Inferenz und Feintuning mit einem Kontextfenster von 128k."
   },
@@ -131,9 +137,15 @@
   "ERNIE-Speed-Pro-128K": {
     "description": "Das neueste von Baidu im Jahr 2024 veröffentlichte hochleistungsfähige Sprachmodell, das überragende allgemeine Fähigkeiten bietet und bessere Ergebnisse als ERNIE Speed erzielt. Es eignet sich als Basis-Modell für Feinabstimmungen, um spezifische Szenarien besser zu bearbeiten, und bietet gleichzeitig hervorragende Inferenzleistung."
   },
+  "FLUX-1.1-pro": {
+    "description": "FLUX.1.1 Pro"
+  },
   "FLUX.1-Kontext-dev": {
     "description": "FLUX.1-Kontext-dev ist ein von Black Forest Labs entwickeltes multimodales Bildgenerierungs- und Bearbeitungsmodell auf Basis der Rectified Flow Transformer-Architektur mit 12 Milliarden Parametern. Es konzentriert sich auf die Generierung, Rekonstruktion, Verbesserung oder Bearbeitung von Bildern unter gegebenen Kontextbedingungen. Das Modell kombiniert die kontrollierbare Generierung von Diffusionsmodellen mit der Kontextmodellierung von Transformern, unterstützt hochwertige Bildausgaben und ist vielseitig einsetzbar für Bildrestaurierung, Bildvervollständigung und visuelle Szenenrekonstruktion."
   },
+  "FLUX.1-Kontext-pro": {
+    "description": "FLUX.1 Kontext [pro]"
+  },
   "FLUX.1-dev": {
     "description": "FLUX.1-dev ist ein von Black Forest Labs entwickeltes Open-Source-multimodales Sprachmodell (Multimodal Language Model, MLLM), das für Bild-Text-Aufgaben optimiert ist und Verständnis sowie Generierung von Bildern und Texten vereint. Es basiert auf fortschrittlichen großen Sprachmodellen wie Mistral-7B und erreicht durch sorgfältig gestaltete visuelle Encoder und mehrstufige Instruktions-Feinabstimmung eine kooperative Verarbeitung von Bild und Text sowie komplexe Aufgabenlogik."
   },
@@ -170,6 +182,9 @@
   "Llama-3.2-90B-Vision-Instruct\t": {
     "description": "Fortgeschrittene Bildschlussfolgerungsfähigkeiten für Anwendungen im Bereich der visuellen Verständigung."
   },
+  "Meta-Llama-3-3-70B-Instruct": {
+    "description": "Llama 3.3 70B: Ein vielseitiges Transformer-Modell, geeignet für Dialog- und Generierungsaufgaben."
+  },
   "Meta-Llama-3.1-405B-Instruct": {
     "description": "Das auf Anweisungen optimierte Textmodell Llama 3.1 wurde für mehrsprachige Dialoganwendungen optimiert und zeigt in vielen verfügbaren Open-Source- und geschlossenen Chat-Modellen in gängigen Branchenbenchmarks hervorragende Leistungen."
   },
@@ -188,6 +203,9 @@
   "Meta-Llama-3.3-70B-Instruct": {
     "description": "Llama 3.3 ist das fortschrittlichste mehrsprachige Open-Source-Sprachmodell der Llama-Serie, das eine Leistung bietet, die mit einem 405B-Modell vergleichbar ist, und das zu extrem niedrigen Kosten. Es basiert auf der Transformer-Architektur und wurde durch überwachte Feinabstimmung (SFT) und verstärkendes Lernen mit menschlichem Feedback (RLHF) in Bezug auf Nützlichkeit und Sicherheit verbessert. Die auf Anweisungen optimierte Version ist speziell für mehrsprachige Dialoge optimiert und übertrifft in mehreren Branchenbenchmarks viele verfügbare Open-Source- und geschlossene Chat-Modelle. Das Wissensdatum endet im Dezember 2023."
   },
+  "Meta-Llama-4-Maverick-17B-128E-Instruct-FP8": {
+    "description": "Llama 4 Maverick: Ein groß angelegtes Modell basierend auf Mixture-of-Experts, das eine effiziente Expertenaktivierungsstrategie bietet, um bei der Inferenz herausragende Leistungen zu erzielen."
+  },
   "MiniMax-M1": {
     "description": "Ein völlig neu entwickeltes Inferenzmodell. Weltweit führend: 80K Denkketten x 1M Eingaben, Leistung auf Augenhöhe mit den besten Modellen im Ausland."
   },
@@ -257,6 +275,9 @@
   "Pro/deepseek-ai/DeepSeek-V3": {
     "description": "DeepSeek-V3 ist ein hybrides Experten (MoE) Sprachmodell mit 6710 Milliarden Parametern, das eine Multi-Head-Latente-Attention (MLA) und DeepSeekMoE-Architektur verwendet, kombiniert mit einer Lastenausgleichsstrategie ohne Hilfskosten, um die Inferenz- und Trainingseffizienz zu optimieren. Durch das Pre-Training auf 14,8 Billionen hochwertigen Tokens und anschließende überwachte Feinabstimmung und verstärktes Lernen übertrifft DeepSeek-V3 in der Leistung andere Open-Source-Modelle und nähert sich führenden geschlossenen Modellen."
   },
+  "Pro/deepseek-ai/DeepSeek-V3.1": {
+    "description": "DeepSeek-V3.1 ist ein hybrides großes Sprachmodell, das von DeepSeek AI veröffentlicht wurde und auf dem Vorgängermodell in vielerlei Hinsicht bedeutende Verbesserungen aufweist. Eine wesentliche Innovation dieses Modells ist die Integration des „Denkmodus“ und des „Nicht-Denkmodus“ in einem System, wobei Nutzer durch Anpassung der Chat-Vorlagen flexibel zwischen den Modi wechseln können, um unterschiedlichen Aufgabenanforderungen gerecht zu werden. Durch spezielles Post-Training wurde die Leistung von V3.1 bei Tool-Aufrufen und Agentenaufgaben deutlich gesteigert, was eine bessere Unterstützung externer Suchwerkzeuge und die Ausführung komplexer mehrstufiger Aufgaben ermöglicht. Das Modell basiert auf DeepSeek-V3.1-Base und wurde durch eine zweistufige Langtext-Erweiterungsmethode nachtrainiert, wodurch das Trainingsdatenvolumen erheblich erhöht wurde und es sich besonders bei der Verarbeitung langer Dokumente und umfangreicher Codes bewährt. Als Open-Source-Modell zeigt DeepSeek-V3.1 in Benchmarks zu Codierung, Mathematik und logischem Denken Fähigkeiten, die mit führenden Closed-Source-Modellen vergleichbar sind. Gleichzeitig senkt seine hybride Expertenarchitektur (MoE) die Inferenzkosten bei gleichzeitiger Beibehaltung einer enormen Modellkapazität."
+  },
   "Pro/moonshotai/Kimi-K2-Instruct": {
     "description": "Kimi K2 ist ein MoE-Architektur-Basis-Modell mit herausragenden Code- und Agentenfähigkeiten, insgesamt 1 Billion Parameter und 32 Milliarden aktivierten Parametern. In Benchmark-Tests zu allgemeinem Wissen, Programmierung, Mathematik und Agentenaufgaben übertrifft das K2-Modell andere führende Open-Source-Modelle."
   },
@@ -332,12 +353,21 @@
   "Qwen/Qwen3-30B-A3B-Instruct-2507": {
     "description": "Qwen3-30B-A3B-Instruct-2507 ist eine aktualisierte Version des Qwen3-30B-A3B im Nicht-Denkmodus. Es handelt sich um ein Mixture-of-Experts (MoE)-Modell mit insgesamt 30,5 Milliarden Parametern und 3,3 Milliarden Aktivierungsparametern. Das Modell wurde in mehreren Bereichen entscheidend verbessert, darunter eine signifikante Steigerung der Befolgung von Anweisungen, logisches Denken, Textverständnis, Mathematik, Wissenschaft, Programmierung und Werkzeugnutzung. Gleichzeitig wurden substanzielle Fortschritte bei der Abdeckung von Langschwanzwissen in mehreren Sprachen erzielt, und es kann besser auf die Präferenzen der Nutzer bei subjektiven und offenen Aufgaben abgestimmt werden, um hilfreichere Antworten und qualitativ hochwertigere Texte zu generieren. Darüber hinaus wurde die Fähigkeit zum Verständnis langer Texte auf 256K erweitert. Dieses Modell unterstützt ausschließlich den Nicht-Denkmodus und generiert keine `<think></think>`-Tags in der Ausgabe."
   },
+  "Qwen/Qwen3-30B-A3B-Thinking-2507": {
+    "description": "Qwen3-30B-A3B-Thinking-2507 ist das neueste Denkmodell der Qwen3‑Serie, veröffentlicht vom Alibaba Tongyi Qianwen‑Team. Als ein Mixture-of-Experts-(MoE)-Modell mit 30,5 Milliarden Gesamtparametern und 3,3 Milliarden aktiven Parametern konzentriert es sich auf die Verbesserung der Bewältigung komplexer Aufgaben. Das Modell zeigt deutliche Leistungssteigerungen in akademischen Benchmarks für logisches Schließen, Mathematik, Naturwissenschaften, Programmierung sowie Aufgaben, die menschliche Fachkenntnisse erfordern. Gleichzeitig wurden seine allgemeinen Fähigkeiten bei der Befolgung von Anweisungen, der Nutzung von Werkzeugen, der Textgenerierung und der Anpassung an menschliche Präferenzen erheblich gestärkt. Das Modell unterstützt nativ ein langes Kontextverständnis von 256K und ist auf bis zu 1 Million Tokens skalierbar. Diese Version ist speziell für den 'Denkmodus' konzipiert und zielt darauf ab, hochkomplexe Aufgaben durch ausführliches, schrittweises Denken zu lösen; auch seine Agent‑Fähigkeiten sind hervorragend."
+  },
   "Qwen/Qwen3-32B": {
     "description": "Qwen3 ist ein neues, leistungsstark verbessertes Modell von Tongyi Qianwen, das in den Bereichen Denken, Allgemeinwissen, Agenten und Mehrsprachigkeit in mehreren Kernfähigkeiten branchenführende Standards erreicht und den Wechsel zwischen Denkmodi unterstützt."
   },
   "Qwen/Qwen3-8B": {
     "description": "Qwen3 ist ein neues, leistungsstark verbessertes Modell von Tongyi Qianwen, das in den Bereichen Denken, Allgemeinwissen, Agenten und Mehrsprachigkeit in mehreren Kernfähigkeiten branchenführende Standards erreicht und den Wechsel zwischen Denkmodi unterstützt."
   },
+  "Qwen/Qwen3-Coder-30B-A3B-Instruct": {
+    "description": "Qwen3-Coder-30B-A3B-Instruct ist ein Code-Modell der Qwen3-Serie, das vom Alibaba-Team Tongyi Qianwen entwickelt wurde. Als schlank optimiertes Modell konzentriert es sich darauf, die Code-Verarbeitungsfähigkeiten zu verbessern, während es hohe Leistung und Effizienz beibehält. Das Modell zeigt unter Open-Source-Modellen deutliche Leistungsvorteile bei komplexen Aufgaben wie agentischem Programmieren (Agentic Coding), automatisierten Browseroperationen und Werkzeugaufrufen. Es unterstützt nativ lange Kontexte mit 256K Tokens und lässt sich auf bis zu 1M Tokens erweitern, sodass es Verständnis- und Verarbeitungsaufgaben auf Ebene ganzer Codebasen besser bewältigen kann. Darüber hinaus bietet das Modell starke Agenten-Codierungsunterstützung für Plattformen wie Qwen Code und CLINE und verfügt über ein speziell entwickeltes Format für Funktionsaufrufe."
+  },
+  "Qwen/Qwen3-Coder-480B-A35B-Instruct": {
+    "description": "Qwen3-Coder-480B-A35B-Instruct wurde von Alibaba veröffentlicht und ist bislang das agentischste Code-Modell. Es ist ein Mixture-of-Experts-(MoE)-Modell mit 480 Milliarden Gesamtparametern und 35 Milliarden aktivierten Parametern, das ein ausgewogenes Verhältnis von Effizienz und Leistung bietet. Das Modell unterstützt nativ eine Kontextlänge von 256K (≈260.000) Token und lässt sich mittels Extrapolationsverfahren wie YaRN auf bis zu 1.000.000 Token erweitern, sodass es große Codebasen und komplexe Programmieraufgaben verarbeiten kann. Qwen3-Coder wurde für agentenbasierte Coding-Workflows entwickelt: Es generiert nicht nur Code, sondern kann auch eigenständig mit Entwicklungswerkzeugen und -umgebungen interagieren, um komplexe Programmierprobleme zu lösen. In mehreren Benchmarks zu Coding- und Agentenaufgaben gehört das Modell zu den Spitzenreitern unter Open-Source-Modellen und erreicht eine Leistungsfähigkeit, die mit führenden Modellen wie Claude Sonnet 4 vergleichbar ist."
+  },
   "Qwen2-72B-Instruct": {
     "description": "Qwen2 ist die neueste Reihe des Qwen-Modells, das 128k Kontext unterstützt. Im Vergleich zu den derzeit besten Open-Source-Modellen übertrifft Qwen2-72B in den Bereichen natürliche Sprachverständnis, Wissen, Code, Mathematik und Mehrsprachigkeit deutlich die führenden Modelle."
   },
@@ -368,6 +398,9 @@
   "Qwen3-235B": {
     "description": "Qwen3-235B-A22B ist ein MoE (Mixture-of-Experts)-Modell, das den „Hybrid-Reasoning-Modus“ einführt und Nutzern nahtloses Umschalten zwischen „Denkmodus“ und „Nicht-Denkmodus“ ermöglicht. Es unterstützt das Verständnis und die Argumentation in 119 Sprachen und Dialekten und verfügt über leistungsstarke Werkzeugaufruffähigkeiten. In umfassenden Benchmark-Tests zu allgemeinen Fähigkeiten, Programmierung und Mathematik, Mehrsprachigkeit, Wissen und Argumentation konkurriert es mit führenden aktuellen Großmodellen auf dem Markt wie DeepSeek R1, OpenAI o1, o3-mini, Grok 3 und Google Gemini 2.5 Pro."
   },
+  "Qwen3-235B-A22B-Instruct-2507-FP8": {
+    "description": "Qwen3 235B A22B Instruct 2507: Ein Modell, optimiert für fortgeschrittene Schlussfolgerungen und Dialoganweisungen, mit einer hybriden Expertenarchitektur, die bei großem Parameterumfang die Inferenz-Effizienz bewahrt."
+  },
   "Qwen3-32B": {
     "description": "Qwen3-32B ist ein dichtes Modell (Dense Model), das den „Hybrid-Reasoning-Modus“ einführt und Nutzern nahtloses Umschalten zwischen „Denkmodus“ und „Nicht-Denkmodus“ ermöglicht. Aufgrund von Verbesserungen in der Modellarchitektur, einer Erweiterung der Trainingsdaten und effizienteren Trainingsmethoden entspricht die Gesamtleistung der von Qwen2.5-72B."
   },
@@ -824,6 +857,9 @@
   "deepseek-ai/DeepSeek-V3": {
     "description": "DeepSeek-V3 ist ein hybrides Expertenmodell (MoE) mit 6710 Milliarden Parametern, das eine Multi-Head-Latent-Attention (MLA) und die DeepSeekMoE-Architektur verwendet, kombiniert mit einer Lastenausgleichsstrategie ohne Hilfskosten, um die Inferenz- und Trainingseffizienz zu optimieren. Durch das Pre-Training auf 14,8 Billionen hochwertigen Tokens und anschließendes überwachten Feintuning und verstärkendes Lernen übertrifft DeepSeek-V3 in der Leistung andere Open-Source-Modelle und nähert sich führenden Closed-Source-Modellen."
   },
+  "deepseek-ai/DeepSeek-V3.1": {
+    "description": "DeepSeek-V3.1 ist ein hybrides großes Sprachmodell, das von DeepSeek AI veröffentlicht wurde und auf dem Vorgängermodell in vielerlei Hinsicht bedeutende Verbesserungen aufweist. Eine wesentliche Innovation dieses Modells ist die Integration des „Denkmodus“ und des „Nicht-Denkmodus“ in einem System, wobei Nutzer durch Anpassung der Chat-Vorlagen flexibel zwischen den Modi wechseln können, um unterschiedlichen Aufgabenanforderungen gerecht zu werden. Durch spezielles Post-Training wurde die Leistung von V3.1 bei Tool-Aufrufen und Agentenaufgaben deutlich gesteigert, was eine bessere Unterstützung externer Suchwerkzeuge und die Ausführung komplexer mehrstufiger Aufgaben ermöglicht. Das Modell basiert auf DeepSeek-V3.1-Base und wurde durch eine zweistufige Langtext-Erweiterungsmethode nachtrainiert, wodurch das Trainingsdatenvolumen erheblich erhöht wurde und es sich besonders bei der Verarbeitung langer Dokumente und umfangreicher Codes bewährt. Als Open-Source-Modell zeigt DeepSeek-V3.1 in Benchmarks zu Codierung, Mathematik und logischem Denken Fähigkeiten, die mit führenden Closed-Source-Modellen vergleichbar sind. Gleichzeitig senkt seine hybride Expertenarchitektur (MoE) die Inferenzkosten bei gleichzeitiger Beibehaltung einer enormen Modellkapazität."
+  },
   "deepseek-ai/deepseek-llm-67b-chat": {
     "description": "DeepSeek 67B ist ein fortschrittliches Modell, das für komplexe Dialoge trainiert wurde."
   },
@@ -894,7 +930,7 @@
     "description": "DeepSeek R1 Vollversion mit 671B Parametern, die Echtzeit-Online-Suche unterstützt und über verbesserte Verständnis- und Generierungsfähigkeiten verfügt."
   },
   "deepseek-reasoner": {
-    "description": "Das von DeepSeek entwickelte Inferenzmodell. Bevor das Modell die endgültige Antwort ausgibt, gibt es zunächst eine Denkprozesskette aus, um die Genauigkeit der endgültigen Antwort zu erhöhen."
+    "description": "DeepSeek V3.1 Denkmodus. Bevor die endgültige Antwort ausgegeben wird, generiert das Modell eine Kette von Überlegungen, um die Genauigkeit der finalen Antwort zu verbessern."
   },
   "deepseek-v2": {
     "description": "DeepSeek V2 ist ein effizientes Mixture-of-Experts-Sprachmodell, das für wirtschaftliche Verarbeitungsanforderungen geeignet ist."
@@ -908,12 +944,18 @@
   "deepseek-v3-0324": {
     "description": "DeepSeek-V3-0324 ist ein MoE-Modell mit 671 Milliarden Parametern, das in den Bereichen Programmierung und technische Fähigkeiten, Kontextverständnis und Verarbeitung langer Texte herausragende Vorteile bietet."
   },
+  "deepseek-v3.1": {
+    "description": "DeepSeek-V3.1 ist ein neu eingeführtes hybrides Inferenzmodell von DeepSeek, das zwei Inferenzmodi unterstützt: Denkmodus und Nicht-Denkmodus. Es ist effizienter im Denkprozess als DeepSeek-R1-0528. Durch Post-Training-Optimierung wurden die Nutzung von Agenten-Tools und die Leistung bei Agentenaufgaben erheblich verbessert. Unterstützt ein Kontextfenster von 128k und eine maximale Ausgabelänge von 64k Tokens."
+  },
   "deepseek/deepseek-chat-v3-0324": {
     "description": "DeepSeek V3 ist ein Experten-Mischmodell mit 685B Parametern und die neueste Iteration der Flaggschiff-Chatmodellreihe des DeepSeek-Teams.\n\nEs erbt das [DeepSeek V3](/deepseek/deepseek-chat-v3) Modell und zeigt hervorragende Leistungen in verschiedenen Aufgaben."
   },
   "deepseek/deepseek-chat-v3-0324:free": {
     "description": "DeepSeek V3 ist ein Experten-Mischmodell mit 685B Parametern und die neueste Iteration der Flaggschiff-Chatmodellreihe des DeepSeek-Teams.\n\nEs erbt das [DeepSeek V3](/deepseek/deepseek-chat-v3) Modell und zeigt hervorragende Leistungen in verschiedenen Aufgaben."
   },
+  "deepseek/deepseek-chat-v3.1": {
+    "description": "DeepSeek-V3.1 ist ein großes hybrides Inferenzmodell, das 128K langen Kontext und effizienten Moduswechsel unterstützt. Es erzielt herausragende Leistung und Geschwindigkeit bei Tool-Aufrufen, Codegenerierung und komplexen Inferenzaufgaben."
+  },
   "deepseek/deepseek-r1": {
     "description": "DeepSeek-R1 hat die Schlussfolgerungsfähigkeiten des Modells erheblich verbessert, selbst bei nur wenigen gekennzeichneten Daten. Bevor das Modell die endgültige Antwort ausgibt, gibt es zunächst eine Denkprozesskette aus, um die Genauigkeit der endgültigen Antwort zu erhöhen."
   },
@@ -1103,12 +1145,27 @@
   "flux-dev": {
     "description": "FLUX.1 [dev] ist ein Open-Source-Gewichtungs- und Feinschlichtungsmodell für nicht-kommerzielle Anwendungen. Es bietet eine Bildqualität und Instruktionsbefolgung ähnlich der professionellen FLUX-Version, jedoch mit höherer Effizienz. Im Vergleich zu Standardmodellen gleicher Größe ist es ressourcenschonender."
   },
+  "flux-kontext-max": {
+    "description": "Modernste kontextbezogene Bildgenerierung und -bearbeitung – verbindet Text und Bilder, um präzise, kohärente Ergebnisse zu erzielen."
+  },
+  "flux-kontext-pro": {
+    "description": "Modernste kontextbezogene Bildgenerierung und -bearbeitung – verbindet Text und Bild zu präzisen, kohärenten Ergebnissen."
+  },
   "flux-kontext/dev": {
     "description": "FLUX.1 Modell, spezialisiert auf Bildbearbeitungsaufgaben, unterstützt Text- und Bildeingaben."
   },
   "flux-merged": {
     "description": "Das FLUX.1-merged Modell kombiniert die tiefgehenden Eigenschaften, die in der Entwicklungsphase von „DEV“ erforscht wurden, mit der hohen Ausführungsgeschwindigkeit von „Schnell“. Dadurch werden sowohl die Leistungsgrenzen des Modells erweitert als auch dessen Anwendungsbereich vergrößert."
   },
+  "flux-pro": {
+    "description": "Premium-kommerzielles KI-Bildgenerierungsmodell — unvergleichliche Bildqualität und vielfältige Ausgabemöglichkeiten."
+  },
+  "flux-pro-1.1": {
+    "description": "Verbessertes professionelles KI-Modell zur Bildgenerierung — bietet herausragende Bildqualität und eine präzise Umsetzung von Eingabeaufforderungen."
+  },
+  "flux-pro-1.1-ultra": {
+    "description": "Ultrahochauflösende KI-Bildgenerierung — unterstützt Ausgaben mit 4 Megapixeln und erstellt hochauflösende Bilder innerhalb von 10 Sekunden."
+  },
   "flux-pro/kontext": {
     "description": "FLUX.1 Kontext [pro] kann Text und Referenzbilder als Eingabe verarbeiten und ermöglicht nahtlose zielgerichtete lokale Bearbeitungen sowie komplexe umfassende Szenenveränderungen."
   },
@@ -1193,6 +1250,12 @@
   "gemini-2.5-flash": {
     "description": "Gemini 2.5 Flash ist Googles kosteneffizientestes Modell und bietet umfassende Funktionen."
   },
+  "gemini-2.5-flash-image-preview": {
+    "description": "Nano Banana ist Googles neuestes, schnellstes und effizientestes natives multimodales Modell, das es Ihnen ermöglicht, Bilder durch Dialog zu generieren und zu bearbeiten."
+  },
+  "gemini-2.5-flash-image-preview:image": {
+    "description": "Nano Banana ist Googles neuestes, schnellstes und effizientestes natives multimodales Modell, das es Ihnen ermöglicht, Bilder durch Dialog zu generieren und zu bearbeiten."
+  },
   "gemini-2.5-flash-lite": {
     "description": "Gemini 2.5 Flash-Lite ist Googles kleinstes und kosteneffizientestes Modell, das speziell für den großflächigen Einsatz entwickelt wurde."
   },
@@ -1295,6 +1358,9 @@
   "glm-4.5-x": {
     "description": "Die Turbo-Version von GLM-4.5, die bei starker Leistung eine Generierungsgeschwindigkeit von bis zu 100 Tokens pro Sekunde erreicht."
   },
+  "glm-4.5v": {
+    "description": "Das neue visuelle Inferenzmodell der nächsten Generation von Zhipu, basierend auf der MOE-Architektur, verfügt über 106B Gesamtparameter und 12B aktivierte Parameter und erzielt in verschiedenen Benchmarks State-of-the-Art‑Ergebnisse (SOTA) unter weltweit vergleichbaren Open‑Source‑multimodalen Modellen. Es deckt gängige Aufgaben wie Bild-, Video- und Dokumentenverständnis sowie GUI‑Aufgaben ab."
+  },
   "glm-4v": {
     "description": "GLM-4V bietet starke Fähigkeiten zur Bildverständnis und -schlussfolgerung und unterstützt eine Vielzahl visueller Aufgaben."
   },
@@ -1331,6 +1397,12 @@
   "google/gemini-2.5-flash": {
     "description": "Gemini 2.5 Flash ist Googles fortschrittlichstes Hauptmodell, speziell entwickelt für anspruchsvolle Aufgaben in den Bereichen logisches Denken, Programmierung, Mathematik und Wissenschaft. Es verfügt über eingebaute \"Denkfähigkeiten\", die es ermöglichen, Antworten mit höherer Genauigkeit und detaillierter Kontextverarbeitung zu liefern.\n\nHinweis: Dieses Modell gibt es in zwei Varianten: mit und ohne Denkfähigkeit. Die Preisgestaltung für die Ausgabe variiert erheblich, je nachdem, ob die Denkfähigkeit aktiviert ist. Wenn Sie die Standardvariante (ohne den Suffix \":thinking\") wählen, vermeidet das Modell ausdrücklich die Erzeugung von Denk-Token.\n\nUm die Denkfähigkeit zu nutzen und Denk-Token zu erhalten, müssen Sie die \":thinking\"-Variante wählen, was zu höheren Kosten für die Denk-Ausgabe führt.\n\nDarüber hinaus kann Gemini 2.5 Flash über den Parameter \"Maximale Tokenanzahl für das Denken\" konfiguriert werden, wie in der Dokumentation beschrieben (https://openrouter.ai/docs/use-cases/reasoning-tokens#max-tokens-for-reasoning)."
   },
+  "google/gemini-2.5-flash-image-preview": {
+    "description": "Gemini 2.5 Flash Experimentelles Modell, unterstützt Bildgenerierung"
+  },
+  "google/gemini-2.5-flash-image-preview:free": {
+    "description": "Gemini 2.5 Flash Experimentelles Modell, unterstützt Bildgenerierung"
+  },
   "google/gemini-2.5-flash-preview": {
     "description": "Gemini 2.5 Flash ist Googles fortschrittlichstes Hauptmodell, das für fortgeschrittenes Denken, Codierung, Mathematik und wissenschaftliche Aufgaben entwickelt wurde. Es enthält die eingebaute Fähigkeit zu \"denken\", was es ihm ermöglicht, Antworten mit höherer Genauigkeit und detaillierter Kontextverarbeitung zu liefern.\n\nHinweis: Dieses Modell hat zwei Varianten: Denken und Nicht-Denken. Die Ausgabepreise variieren erheblich, je nachdem, ob die Denkfähigkeit aktiviert ist oder nicht. Wenn Sie die Standardvariante (ohne den Suffix \":thinking\") wählen, wird das Modell ausdrücklich vermeiden, Denk-Tokens zu generieren.\n\nUm die Denkfähigkeit zu nutzen und Denk-Tokens zu erhalten, müssen Sie die \":thinking\"-Variante wählen, was zu höheren Preisen für Denk-Ausgaben führt.\n\nDarüber hinaus kann Gemini 2.5 Flash über den Parameter \"maximale Tokenanzahl für das Denken\" konfiguriert werden, wie in der Dokumentation beschrieben (https://openrouter.ai/docs/use-cases/reasoning-tokens#max-tokens-for-reasoning)."
   },
@@ -1434,7 +1506,7 @@
     "description": "GPT-4.1 mini bietet ein Gleichgewicht zwischen Intelligenz, Geschwindigkeit und Kosten, was es zu einem attraktiven Modell für viele Anwendungsfälle macht."
   },
   "gpt-4.5-preview": {
-    "description": "Die Forschungs-Vorschau von GPT-4.5, unserem bisher größten und leistungsstärksten GPT-Modell. Es verfügt über umfangreiches Weltwissen und kann die Absichten der Benutzer besser verstehen, was es in kreativen Aufgaben und autonomer Planung herausragend macht. GPT-4.5 akzeptiert Text- und Bild-Eingaben und generiert Textausgaben (einschließlich strukturierter Ausgaben). Es unterstützt wichtige Entwicklerfunktionen wie Funktionsaufrufe, Batch-APIs und Streaming-Ausgaben. In Aufgaben, die kreatives, offenes Denken und Dialog erfordern (wie Schreiben, Lernen oder das Erkunden neuer Ideen), zeigt GPT-4.5 besonders gute Leistungen. Der Wissensstand ist bis Oktober 2023."
+    "description": "GPT-4.5-preview ist das neueste Allzweckmodell, verfügt über fundiertes Weltwissen und ein verbessertes Verständnis der Nutzerintentionen und ist besonders leistungsfähig bei kreativen Aufgaben sowie in der Planung von Agenten. Das Wissen des Modells reicht bis Oktober 2023."
   },
   "gpt-4o": {
     "description": "ChatGPT-4o ist ein dynamisches Modell, das in Echtzeit aktualisiert wird, um die neueste Version zu gewährleisten. Es kombiniert starke Sprachverständnis- und Generierungsfähigkeiten und eignet sich für großangelegte Anwendungsszenarien, einschließlich Kundenservice, Bildung und technische Unterstützung."
@@ -1502,6 +1574,9 @@
   "gpt-oss": {
     "description": "GPT-OSS 20B ist ein von OpenAI veröffentlichtes Open-Source-Sprachmodell, das die MXFP4-Quantisierungstechnologie verwendet und sich für den Einsatz auf High-End-Consumer-GPUs oder Apple Silicon Macs eignet. Dieses Modell zeigt hervorragende Leistungen bei der Dialoggenerierung, Codeerstellung und bei Inferenzaufgaben und unterstützt Funktionsaufrufe sowie die Nutzung von Werkzeugen."
   },
+  "gpt-oss-120b": {
+    "description": "GPT-OSS-120B MXFP4 quantisierte Transformer-Struktur, die auch bei begrenzten Ressourcen starke Leistung beibehält."
+  },
   "gpt-oss:120b": {
     "description": "GPT-OSS 120B ist ein von OpenAI veröffentlichtes großes Open-Source-Sprachmodell, das die MXFP4-Quantisierungstechnologie verwendet und als Flaggschiff-Modell gilt. Es erfordert den Betrieb auf Multi-GPU- oder Hochleistungs-Workstation-Umgebungen und bietet herausragende Leistungen bei komplexen Inferenzaufgaben, Codegenerierung und mehrsprachiger Verarbeitung. Es unterstützt fortgeschrittene Funktionsaufrufe und die Integration von Werkzeugen."
   },
@@ -1529,6 +1604,9 @@
   "grok-4": {
     "description": "Unser neuestes und leistungsstärkstes Flaggschiffmodell, das in der Verarbeitung natürlicher Sprache, mathematischen Berechnungen und logischem Denken herausragende Leistungen erbringt – ein perfekter Allrounder."
   },
+  "grok-code-fast-1": {
+    "description": "Wir freuen uns, grok-code-fast-1 vorzustellen, ein schnelles und kosteneffizientes Inferenzmodell, das sich durch hervorragende Leistung bei der Agentencodierung auszeichnet."
+  },
   "gryphe/mythomax-l2-13b": {
     "description": "MythoMax l2 13B ist ein Sprachmodell, das Kreativität und Intelligenz kombiniert und mehrere führende Modelle integriert."
   },
@@ -1637,9 +1715,18 @@
   "image-01-live": {
     "description": "Bildgenerierungsmodell mit feiner Bilddarstellung, unterstützt Text-zu-Bild und Stil-Einstellungen."
   },
+  "imagen-4.0-fast-generate-001": {
+    "description": "Imagen – Text-zu-Bild-Modellreihe der 4. Generation (Fast-Version)"
+  },
+  "imagen-4.0-generate-001": {
+    "description": "Imagen, Text-zu-Bild-Modellreihe der 4. Generation"
+  },
   "imagen-4.0-generate-preview-06-06": {
     "description": "Imagen 4. Generation Text-zu-Bild Modellserie"
   },
+  "imagen-4.0-ultra-generate-001": {
+    "description": "Imagen, Text-zu-Bild-Modell der 4. Generation (Ultra-Version)"
+  },
   "imagen-4.0-ultra-generate-preview-06-06": {
     "description": "Imagen 4. Generation Text-zu-Bild Modellserie Ultra-Version"
   },
@@ -1679,6 +1766,9 @@
   "kimi-k2-0711-preview": {
     "description": "kimi-k2 ist ein MoE-Architektur-Basis-Modell mit außergewöhnlichen Fähigkeiten in Code und Agentenfunktionen, mit insgesamt 1 Billion Parametern und 32 Milliarden aktiven Parametern. In Benchmark-Tests zu allgemeinem Wissen, Programmierung, Mathematik und Agenten übertrifft das K2-Modell andere führende Open-Source-Modelle."
   },
+  "kimi-k2-turbo-preview": {
+    "description": "kimi-k2 ist ein Basis-Modell mit MoE-Architektur und besonders starken Fähigkeiten im Bereich Code und Agenten. Es verfügt über insgesamt 1T Parameter und 32B aktivierte Parameter. In Benchmark-Tests der wichtigsten Kategorien – allgemeines Wissens-Reasoning, Programmierung, Mathematik und Agenten – übertrifft das K2-Modell die Leistung anderer gängiger Open‑Source‑Modelle."
+  },
   "kimi-latest": {
     "description": "Das Kimi intelligente Assistenzprodukt verwendet das neueste Kimi Großmodell, das möglicherweise noch instabile Funktionen enthält. Es unterstützt die Bildverarbeitung und wählt automatisch das Abrechnungsmodell 8k/32k/128k basierend auf der Länge des angeforderten Kontexts aus."
   },
@@ -1763,6 +1853,9 @@
   "llava:34b": {
     "description": "LLaVA ist ein multimodales Modell, das visuelle Encoder und Vicuna kombiniert und für starke visuelle und sprachliche Verständnisse sorgt."
   },
+  "magistral-medium-latest": {
+    "description": "Magistral Medium 1.1 ist ein fortschrittliches Inferenzmodell, das Mistral AI im Juli 2025 veröffentlicht hat."
+  },
   "mathstral": {
     "description": "MathΣtral ist für wissenschaftliche Forschung und mathematische Schlussfolgerungen konzipiert und bietet effektive Rechenfähigkeiten und Ergebnisinterpretationen."
   },
@@ -2094,7 +2187,7 @@
     "description": "o1-mini ist ein schnelles und kosteneffizientes Inferenzmodell, das für Programmier-, Mathematik- und Wissenschaftsanwendungen entwickelt wurde. Das Modell hat einen Kontext von 128K und einen Wissensstand bis Oktober 2023."
   },
   "o1-preview": {
-    "description": "o1 ist OpenAIs neues Inferenzmodell, das für komplexe Aufgaben geeignet ist, die umfangreiches Allgemeinwissen erfordern. Das Modell hat einen Kontext von 128K und einen Wissensstand bis Oktober 2023."
+    "description": "Konzentriert auf fortgeschrittenes Schlussfolgern und die Lösung komplexer Probleme, einschließlich mathematischer und naturwissenschaftlicher Aufgaben. Sehr gut geeignet für Anwendungen, die ein tiefes Kontextverständnis und autonome Arbeitsabläufe benötigen."
   },
   "o1-pro": {
     "description": "Die o1-Serie wurde durch verstärkendes Lernen trainiert, um vor der Antwort nachzudenken und komplexe Schlussfolgerungen zu ziehen. Das o1-pro Modell nutzt mehr Rechenressourcen für tiefere Überlegungen und liefert dadurch kontinuierlich qualitativ hochwertigere Antworten."
@@ -2213,8 +2306,14 @@
   "qwen-coder-turbo-latest": {
     "description": "Das Tongyi Qianwen Code-Modell."
   },
+  "qwen-flash": {
+    "description": "Die Tongyi-Qianwen-Reihe bietet besonders schnelle und sehr kostengünstige Modelle und eignet sich für einfache Aufgaben."
+  },
   "qwen-image": {
-    "description": "Leistungsstarkes Rohbildmodell vom Qwen-Team mit beeindruckenden Fähigkeiten zur chinesischen Textgenerierung und vielfältigen visuellen Bildstilen."
+    "description": "Qwen-Image ist ein universelles Bildgenerierungsmodell, das zahlreiche Kunststile unterstützt und sich besonders bei der Wiedergabe komplexer Texte auszeichnet, insbesondere bei chinesischen und englischen Schriftzügen. Das Modell unterstützt mehrzeilige Layouts, absatzweises Textgenerieren sowie die präzise Darstellung feiner Details und ermöglicht die Erstellung komplexer Bild-Text-Kombinationen."
+  },
+  "qwen-image-edit": {
+    "description": "Das Qwen-Team hat ein professionelles Modell zur Bildbearbeitung veröffentlicht, das semantische Bearbeitungen und Aussehensbearbeitungen unterstützt. Es kann chinesische und englische Texte präzise bearbeiten und ermöglicht Stiltransformationen, Objektrotationen sowie weitere hochwertige Bildbearbeitungen."
   },
   "qwen-long": {
     "description": "Qwen ist ein groß angelegtes Sprachmodell, das lange Textkontexte unterstützt und Dialogfunktionen für verschiedene Szenarien wie lange Dokumente und mehrere Dokumente bietet."
@@ -2241,7 +2340,7 @@
     "description": "Qwen Plus ist die verbesserte Version des großangelegten Sprachmodells, das Eingaben in verschiedenen Sprachen wie Chinesisch und Englisch unterstützt."
   },
   "qwen-turbo": {
-    "description": "Qwen Turbo ist ein großangelegtes Sprachmodell, das Eingaben in verschiedenen Sprachen wie Chinesisch und Englisch unterstützt."
+    "description": "通义千问 Turbo wird künftig nicht mehr aktualisiert. Es wird empfohlen, auf 通义千问 Flash umzusteigen. 通义千问 ist ein äußerst groß angelegtes Sprachmodell und unterstützt Eingaben in Chinesisch, Englisch und weiteren Sprachen."
   },
   "qwen-vl-chat-v1": {
     "description": "Qwen VL unterstützt flexible Interaktionsmethoden, einschließlich Mehrbild-, Mehrfachfragen und kreativen Fähigkeiten."
@@ -2558,9 +2657,15 @@
   "step-2x-large": {
     "description": "Das neue Generationen-Bildmodell von Step Star konzentriert sich auf Bildgenerierung und kann basierend auf Textbeschreibungen des Nutzers hochwertige Bilder erzeugen. Das neue Modell erzeugt realistischere Bildtexturen und bietet verbesserte Fähigkeiten bei der Erzeugung chinesischer und englischer Schriftzeichen."
   },
+  "step-3": {
+    "description": "Dieses Modell verfügt über eine leistungsfähige visuelle Wahrnehmung und ausgeprägte Fähigkeiten zum komplexen Schlussfolgern. Es kann fachübergreifendes Verständnis komplexer Zusammenhänge, die kombinierte Analyse von mathematischen und visuellen Informationen sowie vielfältige visuelle Analyseaufgaben des Alltags präzise bewältigen."
+  },
   "step-r1-v-mini": {
     "description": "Dieses Modell ist ein leistungsstarkes Schlussfolgerungsmodell mit starker Bildverständnisfähigkeit, das in der Lage ist, Bild- und Textinformationen zu verarbeiten und nach tiefem Denken Textinhalte zu generieren. Es zeigt herausragende Leistungen im Bereich der visuellen Schlussfolgerung und verfügt über erstklassige Fähigkeiten in Mathematik, Programmierung und Textschlussfolgerung. Die Kontextlänge beträgt 100k."
   },
+  "stepfun-ai/step3": {
+    "description": "Step3 ist ein wegweisendes multimodales Inferenzmodell, veröffentlicht von StepFun (阶跃星辰). Es basiert auf einer Mixture-of-Experts-(MoE)-Architektur mit insgesamt 321 Milliarden Parametern und 38 Milliarden Aktivierungsparametern. Das Modell ist als End-to-End-System konzipiert, um die Decodierungskosten zu minimieren und gleichzeitig erstklassige Leistung bei visuell-sprachlicher Inferenz zu bieten. Durch die synergistische Kombination von Multi-Matrix-Factorization-Attention (MFA) und Attention-FFN-Dekopplung (AFD) erzielt Step3 sowohl auf High-End- als auch auf ressourcenbeschränkten Beschleunigern hohe Effizienz. In der Vortrainingsphase verarbeitete Step3 mehr als 20 Billionen Text-Tokens und 4 Billionen multimodale (Bild‑Text) Tokens und deckt dabei über zehn Sprachen ab. Das Modell erzielt in zahlreichen Benchmarks — etwa in Mathematik, Programmierung und Multimodalität — führende Ergebnisse unter den Open‑Source‑Modellen."
+  },
   "taichu_llm": {
     "description": "Das Zīdōng Taichu Sprachmodell verfügt über außergewöhnliche Sprachverständnisfähigkeiten sowie Fähigkeiten in Textgenerierung, Wissensabfrage, Programmierung, mathematischen Berechnungen, logischem Denken, Sentimentanalyse und Textzusammenfassung. Es kombiniert innovativ große Datenvortrainings mit reichhaltigem Wissen aus mehreren Quellen, verfeinert kontinuierlich die Algorithmen und absorbiert ständig neues Wissen aus umfangreichen Textdaten in Bezug auf Vokabular, Struktur, Grammatik und Semantik, um die Leistung des Modells kontinuierlich zu verbessern. Es bietet den Nutzern bequemere Informationen und Dienstleistungen sowie ein intelligenteres Erlebnis."
   },
@@ -2588,18 +2693,9 @@
   "thudm/glm-4-9b-chat": {
     "description": "Die Open-Source-Version des neuesten vortrainierten Modells der GLM-4-Serie, das von Zhizhu AI veröffentlicht wurde."
   },
-  "thudm/glm-4-9b:free": {
-    "description": "GLM-4-9B-0414 ist ein Sprachmodell mit 9 Milliarden Parametern aus der GLM-4-Serie, das von THUDM entwickelt wurde. GLM-4-9B-0414 verwendet die gleichen Verstärkungs- und Ausrichtungsstrategien wie das größere 32B-Modell und erzielt in Bezug auf seine Größe hohe Leistungen, was es für ressourcenbeschränkte Bereitstellungen geeignet macht, die dennoch starke Sprachverständnis- und Generierungsfähigkeiten erfordern."
-  },
   "thudm/glm-z1-32b": {
     "description": "GLM-Z1-32B-0414 ist eine verbesserte Denkvariante von GLM-4-32B, die für tiefgehende Mathematik, Logik und codeorientierte Problemlösungen entwickelt wurde. Es verwendet erweiterte Verstärkungslernen (aufgabenspezifisch und basierend auf allgemeinen Paarpräferenzen), um die Leistung bei komplexen mehrstufigen Aufgaben zu verbessern. Im Vergleich zum Basis-GLM-4-32B-Modell hat Z1 die Fähigkeiten im strukturierten Denken und im formalen Bereich erheblich verbessert.\n\nDieses Modell unterstützt die Durchsetzung von \"Denk\"-Schritten durch Prompt-Engineering und bietet verbesserte Kohärenz für Ausgaben im Langformat. Es ist für Agenten-Workflows optimiert und unterstützt langen Kontext (über YaRN), JSON-Toolaufrufe und feinkörnige Sampling-Konfigurationen für stabiles Denken. Besonders geeignet für Anwendungsfälle, die durchdachtes, mehrstufiges Denken oder formale Ableitungen erfordern."
   },
-  "thudm/glm-z1-32b:free": {
-    "description": "GLM-Z1-32B-0414 ist eine verbesserte Denkvariante von GLM-4-32B, die für tiefgehende Mathematik, Logik und codeorientierte Problemlösungen entwickelt wurde. Es verwendet erweiterte Verstärkungslernen (aufgabenspezifisch und basierend auf allgemeinen Paarpräferenzen), um die Leistung bei komplexen mehrstufigen Aufgaben zu verbessern. Im Vergleich zum Basis-GLM-4-32B-Modell hat Z1 die Fähigkeiten im strukturierten Denken und im formalen Bereich erheblich verbessert.\n\nDieses Modell unterstützt die Durchsetzung von \"Denk\"-Schritten durch Prompt-Engineering und bietet verbesserte Kohärenz für Ausgaben im Langformat. Es ist für Agenten-Workflows optimiert und unterstützt langen Kontext (über YaRN), JSON-Toolaufrufe und feinkörnige Sampling-Konfigurationen für stabiles Denken. Besonders geeignet für Anwendungsfälle, die durchdachtes, mehrstufiges Denken oder formale Ableitungen erfordern."
-  },
-  "thudm/glm-z1-9b:free": {
-    "description": "GLM-Z1-9B-0414 ist ein Sprachmodell mit 9B Parametern aus der GLM-4-Serie, das von THUDM entwickelt wurde. Es verwendet Techniken, die ursprünglich auf das größere GLM-Z1-Modell angewendet wurden, einschließlich erweiterten verstärkten Lernens, paarweiser Rangordnungsausrichtung und Training für inferenzintensive Aufgaben wie Mathematik, Programmierung und Logik. Trotz seiner kleineren Größe zeigt es starke Leistungen bei allgemeinen Inferenzaufgaben und übertrifft viele Open-Source-Modelle in Bezug auf seine Gewichtung."
-  },
   "thudm/glm-z1-rumination-32b": {
     "description": "THUDM: GLM Z1 Rumination 32B ist ein tiefes Inferenzmodell mit 32B Parametern aus der GLM-4-Z1-Serie, das für komplexe, offene Aufgaben optimiert wurde, die langes Nachdenken erfordern. Es basiert auf glm-4-32b-0414 und hat zusätzliche Phasen des verstärkten Lernens und mehrstufige Ausrichtungsstrategien hinzugefügt, die die \"Reflexions\"-Fähigkeit einführen, die darauf abzielt, erweiterte kognitive Prozesse zu simulieren. Dazu gehören iterative Inferenz, mehrstufige Analysen und werkzeuggestützte Arbeitsabläufe wie Suche, Abruf und zitationsbewusste Synthese.\n\nDieses Modell zeigt hervorragende Leistungen in forschungsorientiertem Schreiben, vergleichender Analyse und komplexen Fragen und Antworten. Es unterstützt Funktionsaufrufe für Such- und Navigationsprimitiven (`search`, `click`, `open`, `finish`), sodass es in agentenbasierten Pipelines verwendet werden kann. Reflexionsverhalten wird durch ein mehrstufiges Regelbelohnungssystem und verzögerte Entscheidungsmechanismen geformt und wird an tiefen Forschungsrahmen wie dem internen Ausrichtungsstapel von OpenAI gemessen. Diese Variante eignet sich für Szenarien, die Tiefe statt Geschwindigkeit erfordern."
   },
@@ -2707,5 +2803,8 @@
   },
   "zai-org/GLM-4.5-Air": {
     "description": "GLM-4.5-Air ist ein speziell für Agentenanwendungen entwickeltes Basismodell mit Mixture-of-Experts-Architektur. Es ist tief optimiert für Werkzeugaufrufe, Web-Browsing, Softwareentwicklung und Frontend-Programmierung und unterstützt nahtlos die Integration in Code-Agenten wie Claude Code und Roo Code. GLM-4.5 verwendet einen hybriden Inferenzmodus und ist für komplexe Schlussfolgerungen sowie den Alltagsgebrauch geeignet."
+  },
+  "zai-org/GLM-4.5V": {
+    "description": "GLM-4.5V ist das neueste visuell-sprachliche Modell (VLM), das von Zhipu AI veröffentlicht wurde. Das Modell basiert auf dem Flaggschiff-Textmodell GLM-4.5-Air mit insgesamt 106 Milliarden Parametern und 12 Milliarden Aktivierungsparametern und verwendet eine Mixture-of-Experts-(MoE)-Architektur. Es zielt darauf ab, bei geringeren Inferenzkosten herausragende Leistung zu erzielen. Technisch setzt es die Entwicklungslinie von GLM-4.1V-Thinking fort und führt Innovationen wie die dreidimensionale Rotations-Positionskodierung (3D-RoPE) ein, wodurch die Wahrnehmung und das Schließen über dreidimensionale Raumbeziehungen deutlich verbessert werden. Durch Optimierungen in den Phasen des Pre-Trainings, der überwachten Feinabstimmung und des Reinforcement Learnings ist das Modell in der Lage, verschiedene visuelle Inhalte wie Bilder, Videos und lange Dokumente zu verarbeiten; in 41 öffentlichen multimodalen Benchmarks erreichte es Spitzenwerte unter frei verfügbaren Modellen derselben Klasse. Zudem wurde ein \"Denkmodus\"-Schalter hinzugefügt, der es Nutzern erlaubt, flexibel zwischen schneller Reaktion und tiefgehendem Schlussfolgern zu wählen, um Effizienz und Ergebnisqualität auszubalancieren."
   }
 }

locales/de-DE/providers.json

@@ -11,6 +11,9 @@
   "aihubmix": {
     "description": "AiHubMix bietet über eine einheitliche API-Schnittstelle Zugriff auf verschiedene KI-Modelle."
   },
+  "akashchat": {
+    "description": "Akash ist ein genehmigungsfreier Cloud-Ressourcenmarkt, dessen Preisgestaltung im Vergleich zu herkömmlichen Cloud-Anbietern wettbewerbsfähig ist."
+  },
   "anthropic": {
     "description": "Anthropic ist ein Unternehmen, das sich auf Forschung und Entwicklung im Bereich der künstlichen Intelligenz spezialisiert hat und eine Reihe fortschrittlicher Sprachmodelle anbietet, darunter Claude 3.5 Sonnet, Claude 3 Sonnet, Claude 3 Opus und Claude 3 Haiku. Diese Modelle erreichen ein ideales Gleichgewicht zwischen Intelligenz, Geschwindigkeit und Kosten und sind für eine Vielzahl von Anwendungsszenarien geeignet, von unternehmensweiten Arbeitslasten bis hin zu schnellen Reaktionen. Claude 3.5 Sonnet, als neuestes Modell, hat in mehreren Bewertungen hervorragend abgeschnitten und bietet gleichzeitig ein hohes Preis-Leistungs-Verhältnis."
   },
@@ -26,6 +29,9 @@
   "bedrock": {
     "description": "Bedrock ist ein Service von Amazon AWS, der sich darauf konzentriert, Unternehmen fortschrittliche KI-Sprach- und visuelle Modelle bereitzustellen. Die Modellfamilie umfasst die Claude-Serie von Anthropic, die Llama 3.1-Serie von Meta und mehr, und bietet eine Vielzahl von Optionen von leichtgewichtig bis hochleistungsfähig, die Textgenerierung, Dialoge, Bildverarbeitung und andere Aufgaben unterstützen und für Unternehmensanwendungen unterschiedlicher Größen und Anforderungen geeignet sind."
   },
+  "bfl": {
+    "description": "Ein führendes, an vorderster Front tätiges KI-Forschungslabor, das die visuelle Infrastruktur von morgen gestaltet."
+  },
   "cloudflare": {
     "description": "Führen Sie von serverlosen GPUs betriebene Machine-Learning-Modelle im globalen Netzwerk von Cloudflare aus."
   },
@@ -86,6 +92,12 @@
   "moonshot": {
     "description": "Moonshot ist eine Open-Source-Plattform, die von Beijing Dark Side Technology Co., Ltd. eingeführt wurde und eine Vielzahl von Modellen zur Verarbeitung natürlicher Sprache anbietet, die in vielen Bereichen Anwendung finden, darunter, aber nicht beschränkt auf, Inhaltserstellung, akademische Forschung, intelligente Empfehlungen und medizinische Diagnosen, und unterstützt die Verarbeitung langer Texte und komplexer Generierungsaufgaben."
   },
+  "nebius": {
+    "description": "Nebius bietet globalen KI-Innovatoren durch den Aufbau großflächiger GPU-Cluster und einer vertikal integrierten Cloud-Plattform leistungsstarke Infrastruktur."
+  },
+  "newapi": {
+    "description": "Open-Source-Plattform zur einheitlichen Weiterleitung mehrerer KI-Dienste"
+  },
   "novita": {
     "description": "Novita AI ist eine Plattform, die eine Vielzahl von großen Sprachmodellen und API-Diensten für die KI-Bilderzeugung anbietet, die flexibel, zuverlässig und kosteneffektiv ist. Sie unterstützt die neuesten Open-Source-Modelle wie Llama3 und Mistral und bietet umfassende, benutzerfreundliche und automatisch skalierbare API-Lösungen für die Entwicklung generativer KI-Anwendungen, die für das schnelle Wachstum von KI-Startups geeignet sind."
   },

locales/de-DE/setting.json

@@ -237,6 +237,10 @@
       "title": "Historiennachrichten begrenzen",
       "unlimited": "Unbegrenzte Historiennachrichten"
     },
+    "enableStreaming": {
+      "desc": "Aktivieren Sie den Streaming-Ausgang, um Antworten in Echtzeit anzuzeigen. Wenn deaktiviert, wird nur die vollständige Antwort angezeigt.",
+      "title": "Streaming-Ausgabe aktivieren"
+    },
     "historyCount": {
       "desc": "Anzahl der Nachrichten pro Anfrage (einschließlich der neuesten Fragen und Antworten. Jede Frage und Antwort zählt als 1)",
       "title": "Anzahl der mitgelieferten Nachrichten"

locales/en-US/chat.json

@@ -52,7 +52,11 @@
     "thinking": {
       "title": "Deep Thinking Switch"
     },
-    "title": "Model Extension Features"
+    "title": "Model Extension Features",
+    "urlContext": {
+      "desc": "When enabled, web links will be automatically parsed to retrieve the actual webpage context content",
+      "title": "Extract Webpage Link Content"
+    }
   },
   "history": {
     "title": "The assistant will only remember the last {{count}} messages."
@@ -125,6 +129,7 @@
       "inputWriteCached": "Input cache write",
       "output": "Output",
       "outputAudio": "Audio Output",
+      "outputImage": "Image Output",
       "outputText": "Text Output",
       "outputTitle": "Output Details",
       "reasoning": "Deep Thinking",
@@ -268,7 +273,8 @@
     "clientMode": {
       "actionFiletip": "Upload File",
       "actionTooltip": "Upload",
-      "disabled": "The current model does not support visual recognition and file analysis. Please switch models to use this feature."
+      "disabled": "The current model does not support visual recognition and file analysis. Please switch models to use this feature.",
+      "visionNotSupported": "The current model does not support visual recognition. Please switch to a different model to use this feature."
     },
     "preview": {
       "prepareTasks": "Preparing chunks...",

locales/en-US/common.json

@@ -32,6 +32,7 @@
   "blog": "Product Blog",
   "branching": "Create Subtopic",
   "branchingDisable": "The 'Subtopic' feature is only available in the server version. If you need this feature, please switch to server deployment mode or use LobeChat Cloud.",
+  "branchingRequiresSavedTopic": "Current topic is not saved, please save it first to use subtopic feature",
   "cancel": "Cancel",
   "changelog": "Changelog",
   "clientDB": {
@@ -182,6 +183,13 @@
     "title": "Like Our Product?"
   },
   "fullscreen": "Full Screen Mode",
+  "geminiImageChineseWarning": {
+    "content": "Nano Banana may occasionally fail to generate images when using Chinese. It is recommended to use English for better results.",
+    "continueGenerate": "Continue Generating",
+    "continueSend": "Continue Sending",
+    "doNotShowAgain": "Do Not Show Again",
+    "title": "Chinese Input Notice"
+  },
   "historyRange": "History Range",
   "import": "Import",
   "importData": "Import Data",

locales/en-US/components.json

@@ -73,6 +73,18 @@
   "GoBack": {
     "back": "Back"
   },
+  "HtmlPreview": {
+    "actions": {
+      "download": "Download",
+      "preview": "Preview"
+    },
+    "iframeTitle": "HTML Preview",
+    "mode": {
+      "code": "Code",
+      "preview": "Preview"
+    },
+    "title": "HTML Preview"
+  },
   "ImageUpload": {
     "actions": {
       "changeImage": "Click to change image",
@@ -130,6 +142,12 @@
     },
     "progress": {
       "uploadingWithCount": "{{completed}}/{{total}} uploaded"
+    },
+    "validation": {
+      "fileSizeExceeded": "File size exceeds the maximum allowed",
+      "fileSizeExceededDetail": "{{fileName}} ({{actualSize}}) exceeds the maximum allowed size of {{maxSize}}",
+      "fileSizeExceededMultiple": "{{count}} files exceed the maximum allowed size of {{maxSize}}: {{fileList}}",
+      "imageCountExceeded": "Maximum number of images exceeded"
     }
   },
   "OllamaSetupGuide": {

locales/en-US/error.json

@@ -85,6 +85,17 @@
     "CreateMessageError": "Sorry, the message could not be sent successfully. Please copy the content and try sending it again. This message will not be retained after refreshing the page.",
     "ExceededContextWindow": "The current request content exceeds the length that the model can handle. Please reduce the amount of content and try again.",
     "FreePlanLimit": "You are currently a free user and cannot use this feature. Please upgrade to a paid plan to continue using it.",
+    "GoogleAIBlockReason": {
+      "BLOCKLIST": "Your content contains prohibited terms. Please review and modify your input, then try again.",
+      "IMAGE_SAFETY": "The generated image was blocked for safety reasons. Please try modifying your image request.",
+      "LANGUAGE": "The language you are using is not supported. Please try again in English or another supported language.",
+      "OTHER": "The content was blocked for an unknown reason. Please try rephrasing your request.",
+      "PROHIBITED_CONTENT": "Your request may contain prohibited content. Please adjust your request to comply with the usage guidelines.",
+      "RECITATION": "Your content was blocked due to potential copyright concerns. Please try using original content or rephrase your request.",
+      "SAFETY": "Your content was blocked for safety policy reasons. Please adjust your request to avoid potentially harmful or inappropriate content.",
+      "SPII": "Your content may contain sensitive personally identifiable information (PII). To protect privacy, please remove any sensitive details and try again.",
+      "default": "Content blocked: {{blockReason}}. Please adjust your request and try again."
+    },
     "InsufficientQuota": "Sorry, the quota for this key has been reached. Please check your account balance or increase the key quota and try again.",
     "InvalidAccessCode": "Invalid access code or empty. Please enter the correct access code or add a custom API Key.",
     "InvalidBedrockCredentials": "Bedrock authentication failed. Please check the AccessKeyId/SecretAccessKey and retry.",
@@ -113,18 +124,6 @@
     "PluginServerError": "Plugin server request returned an error. Please check your plugin manifest file, plugin configuration, or server implementation based on the error information below",
     "PluginSettingsInvalid": "This plugin needs to be correctly configured before it can be used. Please check if your configuration is correct",
     "ProviderBizError": "Error requesting {{provider}} service, please troubleshoot or retry based on the following information",
-
-    "GoogleAIBlockReason": {
-      "BLOCKLIST": "Your content contains prohibited terms. Please review and modify your input before retrying.",
-      "IMAGE_SAFETY": "Generated image content was blocked for safety reasons. Please try modifying your image generation request.",
-      "LANGUAGE": "The language you used is not supported. Please try using English or other supported languages.",
-      "OTHER": "Content was blocked for unknown reasons. Please try rephrasing your request or contact technical support.",
-      "PROHIBITED_CONTENT": "Your content may contain prohibited content types. Please adjust your request to ensure it complies with usage guidelines.",
-      "RECITATION": "Your content was blocked due to potential copyright issues. Please try using original content or rephrasing your request.",
-      "SAFETY": "Your content was blocked due to safety policies. Please try adjusting your request to avoid potentially harmful or inappropriate content.",
-      "SPII": "Your content may contain sensitive personally identifiable information. For privacy protection, please remove relevant sensitive information before retrying.",
-      "default": "Content was blocked: {{blockReason}}. Please adjust your request content and try again."
-    },
     "QuotaLimitReached": "We apologize, but the current token usage or number of requests has reached the quota limit for this key. Please increase the quota for this key or try again later.",
     "StreamChunkError": "Error parsing the message chunk of the streaming request. Please check if the current API interface complies with the standard specifications, or contact your API provider for assistance.",
     "SubscriptionKeyMismatch": "We apologize for the inconvenience. Due to a temporary system malfunction, your current subscription usage is inactive. Please click the button below to restore your subscription, or contact us via email for support.",

locales/en-US/modelProvider.json

@@ -153,6 +153,28 @@
     "searchProviders": "Search Providers...",
     "sort": "Custom Sort"
   },
+  "newapi": {
+    "apiKey": {
+      "desc": "API key provided by the New API platform",
+      "placeholder": "New API API Key",
+      "required": "API key is required",
+      "title": "API Key"
+    },
+    "apiUrl": {
+      "desc": "API endpoint for the New API service, usually includes /v1",
+      "title": "API URL"
+    },
+    "enabled": {
+      "title": "Enable New API"
+    },
+    "models": {
+      "batchSelect": "Select models in bulk ({{count}} items)",
+      "fetch": "Fetch model list",
+      "selected": "Selected models",
+      "title": "Available Models"
+    },
+    "title": "New API"
+  },
   "ollama": {
     "checker": {
       "desc": "Test if the proxy address is correctly filled in",
@@ -185,6 +207,10 @@
     }
   },
   "providerModels": {
+    "batchSelect": {
+      "selected": "{{count}} models selected",
+      "title": "Bulk Selection"
+    },
     "config": {
       "aesGcm": "Your key and proxy URL will be encrypted using <1>AES-GCM</1> encryption algorithm",
       "apiKey": {

locales/en-US/models.json

@@ -74,9 +74,15 @@
   "DeepSeek-V3": {
     "description": "DeepSeek-V3 is a MoE model developed in-house by Deep Seek Company. Its performance surpasses that of other open-source models such as Qwen2.5-72B and Llama-3.1-405B in multiple assessments, and it stands on par with the world's top proprietary models like GPT-4o and Claude-3.5-Sonnet."
   },
+  "DeepSeek-V3-1": {
+    "description": "DeepSeek V3.1: Next-generation reasoning model, enhancing complex reasoning and chain-of-thought capabilities, ideal for tasks requiring in-depth analysis."
+  },
   "DeepSeek-V3-Fast": {
     "description": "Model provider: sophnet platform. DeepSeek V3 Fast is the high-TPS ultra-fast version of DeepSeek V3 0324, fully powered without quantization, featuring enhanced coding and mathematical capabilities for faster response!"
   },
+  "DeepSeek-V3.1": {
+    "description": "DeepSeek-V3.1 is a newly launched hybrid reasoning model by DeepSeek, supporting two reasoning modes: thinking and non-thinking. It offers higher thinking efficiency compared to DeepSeek-R1-0528. With post-training optimization, the use of Agent tools and agent task performance have been significantly enhanced."
+  },
   "Doubao-lite-128k": {
     "description": "Doubao-lite offers ultra-fast response times and better cost-effectiveness, providing customers with more flexible options for different scenarios. Supports inference and fine-tuning with a 128k context window."
   },
@@ -131,9 +137,15 @@
   "ERNIE-Speed-Pro-128K": {
     "description": "Baidu's latest self-developed high-performance large language model released in 2024, with outstanding general capabilities, providing better results than ERNIE Speed, suitable as a base model for fine-tuning, effectively addressing specific scenario issues while also exhibiting excellent inference performance."
   },
+  "FLUX-1.1-pro": {
+    "description": "FLUX.1.1 Pro"
+  },
   "FLUX.1-Kontext-dev": {
     "description": "FLUX.1-Kontext-dev is a multimodal image generation and editing model developed by Black Forest Labs based on the Rectified Flow Transformer architecture, featuring 12 billion parameters. It specializes in generating, reconstructing, enhancing, or editing images under given contextual conditions. The model combines the controllable generation advantages of diffusion models with the contextual modeling capabilities of Transformers, supporting high-quality image output and widely applicable to image restoration, completion, and visual scene reconstruction tasks."
   },
+  "FLUX.1-Kontext-pro": {
+    "description": "FLUX.1 Kontext [pro]"
+  },
   "FLUX.1-dev": {
     "description": "FLUX.1-dev is an open-source multimodal language model (MLLM) developed by Black Forest Labs, optimized for vision-and-language tasks by integrating image and text understanding and generation capabilities. Built upon advanced large language models such as Mistral-7B, it achieves vision-language collaborative processing and complex task reasoning through a carefully designed visual encoder and multi-stage instruction fine-tuning."
   },
@@ -170,6 +182,9 @@
   "Llama-3.2-90B-Vision-Instruct\t": {
     "description": "Advanced image reasoning capabilities suitable for visual understanding agent applications."
   },
+  "Meta-Llama-3-3-70B-Instruct": {
+    "description": "Llama 3.3 70B: A versatile Transformer model suitable for conversational and generative tasks."
+  },
   "Meta-Llama-3.1-405B-Instruct": {
     "description": "Llama 3.1 instruction-tuned text model optimized for multilingual dialogue use cases, performing excellently on common industry benchmarks among many available open-source and closed chat models."
   },
@@ -188,6 +203,9 @@
   "Meta-Llama-3.3-70B-Instruct": {
     "description": "Llama 3.3 is the most advanced multilingual open-source large language model in the Llama series, offering performance comparable to a 405B model at a very low cost. Based on the Transformer architecture, it enhances usability and safety through supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF). Its instruction-tuned version is optimized for multilingual dialogue and outperforms many open-source and closed chat models on various industry benchmarks. Knowledge cutoff date is December 2023."
   },
+  "Meta-Llama-4-Maverick-17B-128E-Instruct-FP8": {
+    "description": "Llama 4 Maverick: A large-scale model based on Mixture-of-Experts, offering an efficient expert activation strategy for superior inference performance."
+  },
   "MiniMax-M1": {
     "description": "A brand-new self-developed inference model. Globally leading: 80K reasoning chains x 1M input, performance comparable to top overseas models."
   },
@@ -257,6 +275,9 @@
   "Pro/deepseek-ai/DeepSeek-V3": {
     "description": "DeepSeek-V3 is a mixed expert (MoE) language model with 671 billion parameters, utilizing multi-head latent attention (MLA) and the DeepSeekMoE architecture, combined with a load balancing strategy without auxiliary loss to optimize inference and training efficiency. Pre-trained on 14.8 trillion high-quality tokens and fine-tuned with supervision and reinforcement learning, DeepSeek-V3 outperforms other open-source models and approaches leading closed-source models."
   },
+  "Pro/deepseek-ai/DeepSeek-V3.1": {
+    "description": "DeepSeek-V3.1 is a hybrid large language model released by DeepSeek AI, featuring multiple significant upgrades over its predecessor. A key innovation of this model is the integration of both \"Thinking Mode\" and \"Non-thinking Mode,\" allowing users to flexibly switch between modes by adjusting chat templates to suit different task requirements. Through dedicated post-training optimization, V3.1 significantly enhances performance in tool invocation and Agent tasks, better supporting external search tools and executing complex multi-step tasks. Based on DeepSeek-V3.1-Base, it employs a two-stage long-text extension method to greatly increase training data volume, improving its handling of long documents and extensive code. As an open-source model, DeepSeek-V3.1 demonstrates capabilities comparable to top closed-source models across benchmarks in coding, mathematics, and reasoning. Its Mixture of Experts (MoE) architecture maintains a massive model capacity while effectively reducing inference costs."
+  },
   "Pro/moonshotai/Kimi-K2-Instruct": {
     "description": "Kimi K2 is a MoE architecture base model with exceptional coding and agent capabilities, featuring 1 trillion total parameters and 32 billion activated parameters. In benchmark tests across general knowledge reasoning, programming, mathematics, and agent tasks, the K2 model outperforms other mainstream open-source models."
   },
@@ -332,12 +353,21 @@
   "Qwen/Qwen3-30B-A3B-Instruct-2507": {
     "description": "Qwen3-30B-A3B-Instruct-2507 is an updated version of the Qwen3-30B-A3B non-thinking mode. It is a Mixture of Experts (MoE) model with a total of 30.5 billion parameters and 3.3 billion active parameters. The model features key enhancements across multiple areas, including significant improvements in instruction following, logical reasoning, text comprehension, mathematics, science, coding, and tool usage. Additionally, it has made substantial progress in covering long-tail multilingual knowledge and better aligns with user preferences in subjective and open-ended tasks, enabling it to generate more helpful responses and higher-quality text. Furthermore, its long-text comprehension capability has been extended to 256K tokens. This model supports only the non-thinking mode and does not generate `<think></think>` tags in its output."
   },
+  "Qwen/Qwen3-30B-A3B-Thinking-2507": {
+    "description": "Qwen3-30B-A3B-Thinking-2507 is the latest “thinking” model in the Qwen3 series released by Alibaba’s Tongyi Qianwen team. As a mixture-of-experts (MoE) model with 30.5 billion total parameters and 3.3 billion active parameters, it is designed to improve capabilities for handling complex tasks. The model demonstrates significant performance gains on academic benchmarks requiring logical reasoning, mathematics, science, programming, and domain expertise. At the same time, its general abilities—such as instruction following, tool use, text generation, and alignment with human preferences—have been substantially enhanced. The model natively supports long-context understanding of 256K tokens and can scale up to 1 million tokens. This version is tailored for “thinking mode,” intended to solve highly complex problems through detailed step-by-step reasoning, and it also exhibits strong agent capabilities."
+  },
   "Qwen/Qwen3-32B": {
     "description": "Qwen3 is a next-generation model with significantly enhanced capabilities, achieving industry-leading levels in reasoning, general tasks, agent functions, and multilingual support, with a switchable thinking mode."
   },
   "Qwen/Qwen3-8B": {
     "description": "Qwen3 is a next-generation model with significantly enhanced capabilities, achieving industry-leading levels in reasoning, general tasks, agent functions, and multilingual support, with a switchable thinking mode."
   },
+  "Qwen/Qwen3-Coder-30B-A3B-Instruct": {
+    "description": "Qwen3-Coder-30B-A3B-Instruct is a code model in the Qwen3 series developed by Alibaba's Tongyi Qianwen team. As a streamlined and optimized model, it focuses on enhancing code-handling capabilities while maintaining high performance and efficiency. The model demonstrates notable advantages among open-source models on complex tasks such as agentic coding, automated browser operations, and tool invocation. It natively supports a long context of 256K tokens and can be extended up to 1M tokens, enabling better understanding and processing at the codebase level. Additionally, the model provides robust agentic coding support for platforms like Qwen Code and CLINE, and it employs a dedicated function-calling format."
+  },
+  "Qwen/Qwen3-Coder-480B-A35B-Instruct": {
+    "description": "Qwen3-Coder-480B-A35B-Instruct, released by Alibaba, is the most agentic code model to date. It is a mixture-of-experts (MoE) model with 480 billion total parameters and 35 billion active parameters, striking a balance between efficiency and performance. The model natively supports a 256K (~260k) token context window and can be extended to 1,000,000 tokens through extrapolation methods such as YaRN, enabling it to handle large codebases and complex programming tasks. Qwen3-Coder is designed for agent-style coding workflows: it not only generates code but can autonomously interact with development tools and environments to solve complex programming problems. On multiple benchmarks for coding and agent tasks, this model achieves top-tier results among open-source models, with performance comparable to leading models like Claude Sonnet 4."
+  },
   "Qwen2-72B-Instruct": {
     "description": "Qwen2 is the latest series of the Qwen model, supporting 128k context. Compared to the current best open-source models, Qwen2-72B significantly surpasses leading models in natural language understanding, knowledge, coding, mathematics, and multilingual capabilities."
   },
@@ -368,6 +398,9 @@
   "Qwen3-235B": {
     "description": "Qwen3-235B-A22B is a Mixture of Experts (MoE) model that introduces a \"Hybrid Reasoning Mode,\" allowing users to seamlessly switch between \"Thinking Mode\" and \"Non-Thinking Mode.\" It supports understanding and reasoning in 119 languages and dialects and possesses powerful tool invocation capabilities. In comprehensive benchmarks covering overall ability, coding and mathematics, multilingual proficiency, knowledge, and reasoning, it competes with leading large models on the market such as DeepSeek R1, OpenAI o1, o3-mini, Grok 3, and Google Gemini 2.5 Pro."
   },
+  "Qwen3-235B-A22B-Instruct-2507-FP8": {
+    "description": "Qwen3 235B A22B Instruct 2507: A model optimized for advanced reasoning and dialogue instructions, featuring a mixture-of-experts architecture to maintain inference efficiency at large scale."
+  },
   "Qwen3-32B": {
     "description": "Qwen3-32B is a dense model that introduces a \"Hybrid Reasoning Mode,\" enabling users to seamlessly switch between \"Thinking Mode\" and \"Non-Thinking Mode.\" Thanks to architectural improvements, increased training data, and more efficient training methods, its overall performance is comparable to that of Qwen2.5-72B."
   },
@@ -824,6 +857,9 @@
   "deepseek-ai/DeepSeek-V3": {
     "description": "DeepSeek-V3 is a mixture of experts (MoE) language model with 671 billion parameters, utilizing multi-head latent attention (MLA) and the DeepSeekMoE architecture, combined with a load balancing strategy that does not rely on auxiliary loss, optimizing inference and training efficiency. Pre-trained on 14.8 trillion high-quality tokens and fine-tuned with supervision and reinforcement learning, DeepSeek-V3 outperforms other open-source models and approaches leading closed-source models in performance."
   },
+  "deepseek-ai/DeepSeek-V3.1": {
+    "description": "DeepSeek-V3.1 is a hybrid large language model released by DeepSeek AI, featuring multiple significant upgrades over its predecessor. A key innovation of this model is the integration of both \"Thinking Mode\" and \"Non-thinking Mode,\" allowing users to flexibly switch between modes by adjusting chat templates to suit different task requirements. Through dedicated post-training optimization, V3.1 significantly enhances performance in tool invocation and Agent tasks, better supporting external search tools and executing complex multi-step tasks. Based on DeepSeek-V3.1-Base, it employs a two-stage long-text extension method to greatly increase training data volume, improving its handling of long documents and extensive code. As an open-source model, DeepSeek-V3.1 demonstrates capabilities comparable to top closed-source models across benchmarks in coding, mathematics, and reasoning. Its Mixture of Experts (MoE) architecture maintains a massive model capacity while effectively reducing inference costs."
+  },
   "deepseek-ai/deepseek-llm-67b-chat": {
     "description": "DeepSeek 67B is an advanced model trained for highly complex conversations."
   },
@@ -894,7 +930,7 @@
     "description": "DeepSeek R1 full version, with 671B parameters, supporting real-time online search, offering enhanced understanding and generation capabilities."
   },
   "deepseek-reasoner": {
-    "description": "The reasoning model launched by DeepSeek. Before outputting the final answer, the model first provides a chain of thought to enhance the accuracy of the final response."
+    "description": "DeepSeek V3.1 Thinking Mode. Before outputting the final answer, the model first generates a chain of thought to improve the accuracy of the final response."
   },
   "deepseek-v2": {
     "description": "DeepSeek V2 is an efficient Mixture-of-Experts language model, suitable for cost-effective processing needs."
@@ -908,12 +944,18 @@
   "deepseek-v3-0324": {
     "description": "DeepSeek-V3-0324 is a 671B parameter MoE model, excelling in programming and technical capabilities, contextual understanding, and long text processing."
   },
+  "deepseek-v3.1": {
+    "description": "DeepSeek-V3.1 is a newly launched hybrid reasoning model by DeepSeek, supporting two reasoning modes: thinking and non-thinking. It offers higher thinking efficiency compared to DeepSeek-R1-0528. With post-training optimization, the use of Agent tools and agent task performance have been significantly enhanced. It supports a 128k context window and an output length of up to 64k tokens."
+  },
   "deepseek/deepseek-chat-v3-0324": {
     "description": "DeepSeek V3 is a 685B parameter expert mixture model, the latest iteration in the DeepSeek team's flagship chat model series.\n\nIt inherits from the [DeepSeek V3](/deepseek/deepseek-chat-v3) model and performs excellently across various tasks."
   },
   "deepseek/deepseek-chat-v3-0324:free": {
     "description": "DeepSeek V3 is a 685B parameter expert mixture model, the latest iteration in the DeepSeek team's flagship chat model series.\n\nIt inherits from the [DeepSeek V3](/deepseek/deepseek-chat-v3) model and performs excellently across various tasks."
   },
+  "deepseek/deepseek-chat-v3.1": {
+    "description": "DeepSeek-V3.1 is a large hybrid reasoning model supporting 128K long context and efficient mode switching, delivering outstanding performance and speed in tool invocation, code generation, and complex reasoning tasks."
+  },
   "deepseek/deepseek-r1": {
     "description": "DeepSeek-R1 significantly enhances model reasoning capabilities with minimal labeled data. Before outputting the final answer, the model first provides a chain of thought to improve the accuracy of the final response."
   },
@@ -1103,12 +1145,27 @@
   "flux-dev": {
     "description": "FLUX.1 [dev] is an open-source weight and fine-tuned model for non-commercial applications. It maintains image quality and instruction-following capabilities close to the FLUX professional version while offering higher operational efficiency. Compared to standard models of the same size, it is more resource-efficient."
   },
+  "flux-kontext-max": {
+    "description": "State-of-the-art contextual image generation and editing — combining text and images for precise, coherent results."
+  },
+  "flux-kontext-pro": {
+    "description": "State-of-the-art contextual image generation and editing — combining text and images for precise, coherent results."
+  },
   "flux-kontext/dev": {
     "description": "FLUX.1 model focused on image editing tasks, supporting both text and image inputs."
   },
   "flux-merged": {
     "description": "The FLUX.1-merged model combines the deep features explored during the development phase of “DEV” with the high-speed execution advantages represented by “Schnell.” This integration not only pushes the model's performance boundaries but also broadens its application scope."
   },
+  "flux-pro": {
+    "description": "A top-tier commercial AI image generation model — delivering unparalleled image quality and a wide variety of outputs."
+  },
+  "flux-pro-1.1": {
+    "description": "Upgraded professional-grade AI image generation model — delivers outstanding image quality and precise adherence to prompts."
+  },
+  "flux-pro-1.1-ultra": {
+    "description": "Ultra-high-resolution AI image generation — supports up to 4-megapixel output, producing ultra-high-definition images in under 10 seconds."
+  },
   "flux-pro/kontext": {
     "description": "FLUX.1 Kontext [pro] can process text and reference images as input, seamlessly enabling targeted local edits and complex overall scene transformations."
   },
@@ -1193,6 +1250,12 @@
   "gemini-2.5-flash": {
     "description": "Gemini 2.5 Flash is Google's most cost-effective model, offering comprehensive capabilities."
   },
+  "gemini-2.5-flash-image-preview": {
+    "description": "Nano Banana is Google's latest, fastest, and most efficient native multimodal model, enabling you to generate and edit images through conversation."
+  },
+  "gemini-2.5-flash-image-preview:image": {
+    "description": "Nano Banana is Google's latest, fastest, and most efficient native multimodal model, enabling you to generate and edit images through conversation."
+  },
   "gemini-2.5-flash-lite": {
     "description": "Gemini 2.5 Flash-Lite is Google's smallest and most cost-effective model, designed for large-scale use."
   },
@@ -1295,6 +1358,9 @@
   "glm-4.5-x": {
     "description": "The high-speed version of GLM-4.5, combining strong performance with generation speeds up to 100 tokens per second."
   },
+  "glm-4.5v": {
+    "description": "Zhipu's next-generation visual reasoning model is built on a Mixture-of-Experts (MoE) architecture. With 106B total parameters and 12B activated parameters, it achieves state-of-the-art performance among open-source multimodal models of similar scale across various benchmarks, supporting common tasks such as image, video, document understanding, and GUI-related tasks."
+  },
   "glm-4v": {
     "description": "GLM-4V provides strong image understanding and reasoning capabilities, supporting various visual tasks."
   },
@@ -1331,6 +1397,12 @@
   "google/gemini-2.5-flash": {
     "description": "Gemini 2.5 Flash is Google's most advanced flagship model, designed for advanced reasoning, coding, mathematics, and scientific tasks. It features built-in \"thinking\" capabilities, enabling it to provide responses with higher accuracy and more nuanced contextual understanding.\n\nNote: This model has two variants: thinking and non-thinking. Output pricing varies significantly depending on whether the thinking capability is activated. If you choose the standard variant (without the \":thinking\" suffix), the model will explicitly avoid generating thinking tokens.\n\nTo leverage the thinking capability and receive thinking tokens, you must select the \":thinking\" variant, which incurs higher pricing for thinking outputs.\n\nAdditionally, Gemini 2.5 Flash can be configured via the \"max tokens for reasoning\" parameter, as detailed in the documentation (https://openrouter.ai/docs/use-cases/reasoning-tokens#max-tokens-for-reasoning)."
   },
+  "google/gemini-2.5-flash-image-preview": {
+    "description": "Gemini 2.5 Flash experimental model, supporting image generation."
+  },
+  "google/gemini-2.5-flash-image-preview:free": {
+    "description": "Gemini 2.5 Flash experimental model, supporting image generation."
+  },
   "google/gemini-2.5-flash-preview": {
     "description": "Gemini 2.5 Flash is Google's most advanced flagship model, designed for advanced reasoning, coding, mathematics, and scientific tasks. It includes built-in 'thinking' capabilities that allow it to provide responses with higher accuracy and detailed context handling.\n\nNote: This model has two variants: thinking and non-thinking. Output pricing varies significantly based on whether the thinking capability is activated. If you choose the standard variant (without the ':thinking' suffix), the model will explicitly avoid generating thinking tokens.\n\nTo leverage the thinking capability and receive thinking tokens, you must select the ':thinking' variant, which will incur higher thinking output pricing.\n\nAdditionally, Gemini 2.5 Flash can be configured via the 'maximum tokens for reasoning' parameter, as described in the documentation (https://openrouter.ai/docs/use-cases/reasoning-tokens#max-tokens-for-reasoning)."
   },
@@ -1434,7 +1506,7 @@
     "description": "GPT-4.1 nano provides a balance of intelligence, speed, and cost, making it an appealing model for numerous applications."
   },
   "gpt-4.5-preview": {
-    "description": "The research preview of GPT-4.5, our largest and most powerful GPT model to date. It possesses extensive world knowledge and better understands user intent, excelling in creative tasks and autonomous planning. GPT-4.5 accepts both text and image inputs and generates text outputs (including structured outputs). It supports key developer features such as function calling, batch API, and streaming output. GPT-4.5 particularly shines in tasks that require creativity, open-ended thinking, and dialogue, such as writing, learning, or exploring new ideas. Knowledge cutoff date is October 2023."
+    "description": "GPT-4.5-preview is the latest general-purpose model, offering extensive world knowledge and an improved understanding of user intent. It excels at creative tasks and agent-style planning. The model's knowledge cutoff is October 2023."
   },
   "gpt-4o": {
     "description": "ChatGPT-4o is a dynamic model that updates in real-time to stay current with the latest version. It combines powerful language understanding and generation capabilities, making it suitable for large-scale applications, including customer service, education, and technical support."
@@ -1502,6 +1574,9 @@
   "gpt-oss": {
     "description": "GPT-OSS 20B is an open-source large language model released by OpenAI, utilizing MXFP4 quantization technology. It is suitable for running on high-end consumer GPUs or Apple Silicon Macs. This model excels in dialogue generation, code writing, and reasoning tasks, supporting function calls and tool usage."
   },
+  "gpt-oss-120b": {
+    "description": "GPT-OSS-120B MXFP4 quantized Transformer architecture, delivering strong performance even under resource constraints."
+  },
   "gpt-oss:120b": {
     "description": "GPT-OSS 120B is a large open-source language model released by OpenAI, employing MXFP4 quantization technology as a flagship model. It requires multi-GPU or high-performance workstation environments to operate and delivers outstanding performance in complex reasoning, code generation, and multilingual processing, supporting advanced function calls and tool integration."
   },
@@ -1529,6 +1604,9 @@
   "grok-4": {
     "description": "Our latest and most powerful flagship model, excelling in natural language processing, mathematical computation, and reasoning — a perfect all-rounder."
   },
+  "grok-code-fast-1": {
+    "description": "We are excited to introduce grok-code-fast-1, a fast and cost-effective inference model that excels in agent coding."
+  },
   "gryphe/mythomax-l2-13b": {
     "description": "MythoMax l2 13B is a language model that combines creativity and intelligence by merging multiple top models."
   },
@@ -1637,9 +1715,18 @@
   "image-01-live": {
     "description": "An image generation model with delicate visual performance, supporting text-to-image generation and style setting."
   },
+  "imagen-4.0-fast-generate-001": {
+    "description": "Imagen 4th-generation text-to-image model, Fast version"
+  },
+  "imagen-4.0-generate-001": {
+    "description": "Imagen 4th-generation text-to-image model series"
+  },
   "imagen-4.0-generate-preview-06-06": {
     "description": "Imagen 4th generation text-to-image model series"
   },
+  "imagen-4.0-ultra-generate-001": {
+    "description": "Imagen 4th-generation text-to-image model, Ultra version"
+  },
   "imagen-4.0-ultra-generate-preview-06-06": {
     "description": "Imagen 4th generation text-to-image model series Ultra version"
   },
@@ -1679,6 +1766,9 @@
   "kimi-k2-0711-preview": {
     "description": "kimi-k2 is a MoE architecture base model with powerful coding and agent capabilities, totaling 1 trillion parameters with 32 billion active parameters. In benchmark tests across general knowledge reasoning, programming, mathematics, and agent tasks, the K2 model outperforms other mainstream open-source models."
   },
+  "kimi-k2-turbo-preview": {
+    "description": "Kimi-K2 is a Mixture-of-Experts (MoE) foundation model with exceptional coding and agent capabilities, featuring 1T total parameters and 32B activated parameters. In benchmark evaluations across core categories — general knowledge reasoning, programming, mathematics, and agent tasks — the K2 model outperforms other leading open-source models."
+  },
   "kimi-latest": {
     "description": "The Kimi Smart Assistant product uses the latest Kimi large model, which may include features that are not yet stable. It supports image understanding and will automatically select the 8k/32k/128k model as the billing model based on the length of the request context."
   },
@@ -1763,6 +1853,9 @@
   "llava:34b": {
     "description": "LLaVA is a multimodal model that combines a visual encoder with Vicuna for powerful visual and language understanding."
   },
+  "magistral-medium-latest": {
+    "description": "Magistral Medium 1.1 is a state-of-the-art inference model released by Mistral AI in July 2025."
+  },
   "mathstral": {
     "description": "MathΣtral is designed for scientific research and mathematical reasoning, providing effective computational capabilities and result interpretation."
   },
@@ -2094,7 +2187,7 @@
     "description": "o1-mini is a fast and cost-effective reasoning model designed for programming, mathematics, and scientific applications. This model features a 128K context and has a knowledge cutoff date of October 2023."
   },
   "o1-preview": {
-    "description": "o1 is OpenAI's new reasoning model, suitable for complex tasks that require extensive general knowledge. This model features a 128K context and has a knowledge cutoff date of October 2023."
+    "description": "Focused on advanced reasoning and solving complex problems, including mathematical and scientific tasks. Ideal for applications that require deep contextual understanding and autonomous workflows."
   },
   "o1-pro": {
     "description": "The o1 series models are trained with reinforcement learning to think before answering and perform complex reasoning tasks. The o1-pro model uses more computational resources for deeper thinking, consistently delivering higher-quality responses."
@@ -2213,8 +2306,14 @@
   "qwen-coder-turbo-latest": {
     "description": "The Tongyi Qianwen Coder model."
   },
+  "qwen-flash": {
+    "description": "The Tongyi Qianwen Flash series offers the fastest, most cost-effective models, suitable for simple tasks."
+  },
   "qwen-image": {
-    "description": "A powerful raw image model from the Qwen team, featuring impressive Chinese text generation capabilities and diverse visual styles."
+    "description": "Qwen-Image is a general-purpose image generation model that supports a wide range of artistic styles and is particularly adept at rendering complex text, especially Chinese and English. The model supports multi-line layouts, paragraph-level text generation, and fine-grained detail rendering, enabling complex mixed text-and-image layout designs."
+  },
+  "qwen-image-edit": {
+    "description": "A professional image-editing model released by the Qwen team, supporting semantic editing and appearance editing. It can precisely edit Chinese and English text and perform high-quality image edits such as style transfer and object rotation."
   },
   "qwen-long": {
     "description": "Qwen is a large-scale language model that supports long text contexts and dialogue capabilities based on long documents and multiple documents."
@@ -2241,7 +2340,7 @@
     "description": "Qwen Plus is an enhanced large-scale language model supporting input in various languages including Chinese and English."
   },
   "qwen-turbo": {
-    "description": "Qwen Turbo is a large-scale language model supporting input in various languages including Chinese and English."
+    "description": "Tongyi Qianwen Turbo will no longer receive updates; it is recommended to switch to Tongyi Qianwen Flash. Tongyi Qianwen is an ultra-large language model that supports input in Chinese, English, and other languages."
   },
   "qwen-vl-chat-v1": {
     "description": "Qwen VL supports flexible interaction methods, including multi-image, multi-turn Q&A, and creative capabilities."
@@ -2558,9 +2657,15 @@
   "step-2x-large": {
     "description": "Step Star next-generation image generation model, focusing on image generation tasks. It can generate high-quality images based on user-provided text descriptions. The new model produces more realistic textures and stronger Chinese and English text generation capabilities."
   },
+  "step-3": {
+    "description": "This model has powerful visual perception and advanced reasoning capabilities. It can accurately handle complex cross-domain knowledge comprehension, perform integrated analysis of mathematical and visual information, and solve a wide range of visual analysis tasks encountered in everyday life."
+  },
   "step-r1-v-mini": {
     "description": "This model is a powerful reasoning model with strong image understanding capabilities, able to process both image and text information, generating text content after deep reasoning. It excels in visual reasoning while also possessing first-tier capabilities in mathematics, coding, and text reasoning. The context length is 100k."
   },
+  "stepfun-ai/step3": {
+    "description": "Step3 is a cutting-edge multimodal reasoning model released by StepFun. It is built on a mixture-of-experts (MoE) architecture with 321B total parameters and 38B active parameters. The model adopts an end-to-end design to minimize decoding cost while delivering top-tier performance in visual-language reasoning. Through the combined design of Multi-Matrix Factorized Attention (MFA) and Attention-FFN Decoupling (AFD), Step3 maintains exceptional efficiency on both high-end and low-end accelerators. During pretraining, Step3 processed over 20 trillion text tokens and 4 trillion image-text mixed tokens, covering more than a dozen languages. The model achieves leading performance among open-source models across benchmarks in mathematics, code, and multimodal tasks."
+  },
   "taichu_llm": {
     "description": "The ZD Taichu language model possesses strong language understanding capabilities and excels in text creation, knowledge Q&A, code programming, mathematical calculations, logical reasoning, sentiment analysis, and text summarization. It innovatively combines large-scale pre-training with rich knowledge from multiple sources, continuously refining algorithmic techniques and absorbing new knowledge in vocabulary, structure, grammar, and semantics from vast text data, resulting in an evolving model performance. It provides users with more convenient information and services, as well as a more intelligent experience."
   },
@@ -2588,18 +2693,9 @@
   "thudm/glm-4-9b-chat": {
     "description": "The open-source version of the latest generation pre-trained model from the GLM-4 series released by Zhiyuan AI."
   },
-  "thudm/glm-4-9b:free": {
-    "description": "GLM-4-9B-0414 is a 9 billion parameter language model in the GLM-4 series developed by THUDM. GLM-4-9B-0414 is trained using the same reinforcement learning and alignment strategies as its larger 32B counterpart, achieving high performance relative to its scale, making it suitable for resource-constrained deployments that still require strong language understanding and generation capabilities."
-  },
   "thudm/glm-z1-32b": {
     "description": "GLM-Z1-32B-0414 is an enhanced reasoning variant of GLM-4-32B, built for deep mathematics, logic, and code-oriented problem solving. It applies extended reinforcement learning (task-specific and based on general pairwise preferences) to improve performance on complex multi-step tasks. Compared to the base GLM-4-32B model, Z1 significantly enhances capabilities in structured reasoning and formal domains.\n\nThis model supports enforcing 'thinking' steps through prompt engineering and provides improved coherence for long-format outputs. It is optimized for agent workflows and supports long context (via YaRN), JSON tool calls, and fine-grained sampling configurations for stable reasoning. It is ideal for use cases requiring thoughtful, multi-step reasoning or formal derivation."
   },
-  "thudm/glm-z1-32b:free": {
-    "description": "GLM-Z1-32B-0414 is an enhanced reasoning variant of GLM-4-32B, built for deep mathematics, logic, and code-oriented problem solving. It applies extended reinforcement learning (task-specific and based on general pairwise preferences) to improve performance on complex multi-step tasks. Compared to the base GLM-4-32B model, Z1 significantly enhances capabilities in structured reasoning and formal domains.\n\nThis model supports enforcing 'thinking' steps through prompt engineering and provides improved coherence for long-format outputs. It is optimized for agent workflows and supports long context (via YaRN), JSON tool calls, and fine-grained sampling configurations for stable reasoning. It is ideal for use cases requiring thoughtful, multi-step reasoning or formal derivation."
-  },
-  "thudm/glm-z1-9b:free": {
-    "description": "GLM-Z1-9B-0414 is a 9 billion parameter language model in the GLM-4 series developed by THUDM. It employs techniques initially applied to the larger GLM-Z1 model, including extended reinforcement learning, pairwise ranking alignment, and training for reasoning-intensive tasks such as mathematics, coding, and logic. Despite its smaller scale, it demonstrates strong performance on general reasoning tasks and outperforms many open-source models at its weight level."
-  },
   "thudm/glm-z1-rumination-32b": {
     "description": "THUDM: GLM Z1 Rumination 32B is a deep reasoning model with 32 billion parameters in the GLM-4-Z1 series, optimized for complex, open-ended tasks that require prolonged thought. It builds upon glm-4-32b-0414, adding additional reinforcement learning stages and multi-stage alignment strategies, introducing a 'rumination' capability designed to simulate extended cognitive processing. This includes iterative reasoning, multi-hop analysis, and tool-enhanced workflows such as search, retrieval, and citation-aware synthesis.\n\nThe model excels in research-style writing, comparative analysis, and complex question answering. It supports function calls for search and navigation primitives (`search`, `click`, `open`, `finish`), allowing it to be used in agent-based pipelines. The rumination behavior is shaped by rule-based rewards and a delayed decision-making mechanism, controlled by multi-round cycles, benchmarked against deep research frameworks like OpenAI's internal alignment stack. This variant is suitable for scenarios requiring depth over speed."
   },
@@ -2707,5 +2803,8 @@
   },
   "zai-org/GLM-4.5-Air": {
     "description": "GLM-4.5-Air is a foundational model designed specifically for agent applications, using a Mixture-of-Experts (MoE) architecture. It is deeply optimized for tool invocation, web browsing, software engineering, and front-end programming, supporting seamless integration with code agents like Claude Code and Roo Code. GLM-4.5 employs a hybrid inference mode, adaptable to complex reasoning and everyday use scenarios."
+  },
+  "zai-org/GLM-4.5V": {
+    "description": "GLM-4.5V is the latest-generation vision-language model (VLM) released by Zhipu AI. It is built on the flagship text model GLM-4.5-Air, which has 106B total parameters and 12B active parameters, and adopts a Mixture-of-Experts (MoE) architecture to deliver outstanding performance at reduced inference cost. Technically, GLM-4.5V continues the trajectory of GLM-4.1V-Thinking and introduces innovations such as three-dimensional rotary position encoding (3D-RoPE), significantly improving perception and reasoning of three-dimensional spatial relationships. Through optimizations across pretraining, supervised fine-tuning, and reinforcement learning stages, the model can handle a wide range of visual content including images, video, and long documents, and has achieved top-tier performance among comparable open-source models across 41 public multimodal benchmarks. The model also adds a \"Thinking Mode\" toggle that lets users flexibly choose between fast responses and deep reasoning to balance efficiency and effectiveness."
   }
 }