开源 & 自托管

你的 AI,你的规则

v0.10.0 最新动态
发布说明 查看全部 →
✨ New Features
  • Added a bundled llama.cpp completion provider profile for qwen3-35b-a3b, including configurable endpoint, sampling, token, penalty, and streaming defaults
  • Added an Ollama Cloud category in the Models library and model selector, backed by Ollama's dedicated cloud listing
  • Added custom accent selection in Appearance with violet, blue, cyan, teal, emerald, amber, rose, slate, and custom color options
  • Added runtime-generated accent shade scales so custom accents can drive buttons, links, focus rings, sliders, progress bars, selected states, and glow shadows
🔧 Improvements
  • Reworked Tailwind theme tokens to use CSS variables for primary and accent, with generated custom palettes meeting white-text contrast targets
  • Refined light and dark neutral palettes to reduce glare while preserving contrast across app backgrounds, panels, inputs, and message surfaces
  • Persisted accent preferences through frontend state, backend preferences, import/export defaults, local rehydration, logout cleanup, and demo mode
  • Added accent translation keys across all supported locales and verified locale key parity against English
  • Expanded demo-mode preference APIs so Appearance, generation, embedding, system-message, and default-model settings can be exercised without a backend connection
  • Standardized the Libre WebUI wordmark through a shared Logo component across loading, login, setup, sidebar, and settings surfaces
  • Replaced fixed violet styling in persona adaptive-learning controls with active-accent tokens
  • Recolored Ollama library and Cloud badges from cyan to the design-system info-blue treatment
  • Reduced noisy debug logging across optional auth, chat, document search, plugin lookup, plugin deletion, TTS, image generation, and WebSocket auth paths
  • Returned package version in auth system info for UI/version display
  • Updated Dependabot targeting to main and ignored incompatible Electron and ESLint major bumps until their upstream compatibility gaps are resolved
🐛 Bug Fixes
  • Fixed a critical app-wide Too many re-renders crash in SettingsModal by memoizing image-generation model/plugin arrays
  • Fixed Ollama Cloud model pulls by tagging cloud library results with :cloud before install/run actions
  • Fixed Ollama Cloud generation failures by stripping unsupported non-positive num_predict values before cloud requests
  • Fixed cloud model pull functionality
  • Fixed auto-title generation for persona/current-model workflows by resolving the real running model before sending title prompts
  • Quietly handles expired or invalid JWTs in auth and WebSocket flows without dumping routine stack traces
  • Kept the lint toolchain on ESLint 9 for compatibility with the current React plugin stack
📚 Documentation
  • Expanded DESIGN.md with custom accent presets, generated shade-scale behavior, token usage rules, light/dark neutral guidance, and contrast requirements
📦 Dependencies
  • Bumped package versions to 0.10.0
  • Updated Electron to 41.7.1 and Electron Builder to 26.15.2
  • Bumped frontend/backend dependencies including React Query, i18next, KaTeX, lucide-react, React Router, Vite, pdfjs-dist, TypeScript tooling, Tailwind Merge, Zustand, Express, Multer, Better SQLite3, and related lockfile entries
  • Remediated npm security advisories through dependency and lockfile updates, including qs, tmp, ip-address, brace-expansion, ws, and related transitive packages

隐私优先的 AI 聊天界面。使用 Ollama 本地运行,或连接 OpenAI、Anthropic 等 9+ 服务商。 零遥测。零追踪。

npx libre-webui

需要 Node.js 18+ 和 Ollama (用于本地 AI)

Libre WebUI 界面
Libre WebUI 界面
🔒 零遥测
🏠 自托管
📜 Apache 2.0
🔌 插件系统

你需要的一切

尊重隐私的完整 AI 聊天解决方案

🤖

本地 & 云端 AI

使用 Ollama 本地运行模型,或连接 OpenAI、Anthropic、Groq、Gemini、Mistral 等。选择权在你。

📄

文档对话 (RAG)

上传 PDF、文档和文本文件。通过语义搜索和向量嵌入询问文档相关问题。

🎨

交互式工件

直接在聊天中渲染 HTML、SVG 和 React 组件。支持全屏模式实时预览。

🔐

AES-256 加密

为所有数据提供企业级加密。聊天记录、文档和设置均加密存储。

🎭

自定义角色

创建具有独特行为和系统提示的 AI 个性。支持 JSON 导入/导出角色。

🔊

文字转语音

使用多种语音选项收听 AI 回复。支持浏览器 TTS 和 ElevenLabs 集成。

🖼️

图像生成

使用 ComfyUI 和 Flux 模型生成图像。多种尺寸、质量预设,无缝集成到聊天中。

⚙️

插件变量

每个插件可配置的设置,如温度、端点和令牌。敏感值会加密存储。

⌨️

键盘快捷键

为高级用户提供 VS Code 风格快捷键。通过键盘导航、切换设置、控制一切。

👥

多用户支持

支持 SSO 的基于角色的访问控制。内置 GitHub 和 Hugging Face OAuth。

连接任何服务商

一个界面,无限可能

Ollama
本地模型
OpenAI
GPT-4o, o1, o3
Anthropic
Claude 4, Opus
Groq
Llama, Mixtral
Google
Gemini Pro
Mistral
Mistral Large
OpenRouter
400+ 模型
HuggingFace
1M+ models
+ Custom
任何 OpenAI 兼容 API

几秒钟即可开始

选择你喜欢的安装方式

推荐

npx(一条命令)

npx libre-webui

即时运行。无需安装。

npm(全局安装)

npm install -g libre-webuilibre-webui

安装一次,随处运行。

Docker

docker run -p 8080:8080 libre-webui/libre-webui

容器化部署。

创建自定义插件

通过简单的 JSON 文件连接任何 OpenAI 兼容 LLM

可用插件

Libre WebUI 仓库的官方插件。点击查看或下载。

从 GitHub 加载插件...
📄 custom-model.json
{
  "id": "custom-model",
  "name": "Custom Model",
  "type": "completion",
  "endpoint": "http://localhost:8000/v1/chat/completions",
  "auth": {
    "header": "Authorization",
    "prefix": "Bearer ",
    "key_env": "CUSTOM_MODEL_API_KEY"
  },
  "model_map": [
    "my-fine-tuned-llama"
  ],
  "variables": [
    {
      "name": "temperature",
      "type": "number",
      "label": "Temperature",
      "default": 0.7,
      "min": 0, "max": 2
    }
  ]
}

创建你自己的插件

1

启动 LLM 服务器

运行任何 OpenAI 兼容服务器:llama.cpp、vLLM、Ollama 或自定义 FastAPI 服务器。

2

创建插件 JSON

在简单的 JSON 文件中定义端点、认证和可用模型。

3

上传到 Libre WebUI

进入设置 > 服务商,上传插件并输入 API 密钥。

4

开始聊天

自定义模型将出现在模型选择器中。完全隐私,完全掌控。

插件字段参考

id 唯一标识符(小写,可用连字符)
name UI 中显示的名称
type 聊天用 "completion",语音合成用 "tts"
endpoint API URL(例如:/v1/chat/completions)
auth.header 认证头名称(Authorization、x-api-key)
auth.prefix 密钥前缀("Bearer " 或空)
auth.key_env API 密钥的环境变量
model_map 可用模型标识符数组
variables 可配置的设置(温度、端点等)

准备好掌控你的 AI 了吗?

加入数千名重视隐私和掌控的用户。