Show HN: SelfClaw – Self-hosted multi-model AI chat with OpenAI-compatible API

A developer released SelfClaw, an MIT-licensed self-hosted multi-model AI chat system that aggregates DeepSeek, NVIDIA, SiliconFlow, ZenMux, Zhipu, Kimi, MiMo and Mistral behind an OpenAI-compatible FastAPI gateway. SelfClaw adds automatic failover polling — two retries on temporary errors, a 24-hour blacklist for permanent failures and a 3-minute cooldown when all providers are down — plus a CCR closed-loop agent with tool calling, memory, reflection and strategy evolution, file parsing for PDF, Word and Excel, Bing and Baidu search, and edge-tts voice output. The repo also ships Self-IDE, a local Python service on default port 19898 offering an OpenAI-compatible proxy, goal engine, subagent orchestration and cost meter, with a thin VS Code extension as its shell.

多模型聚合、自动降级、CCR 闭环智能体的 AI 对话系统 - 多模型聚合 — 支持 DeepSeek、NVIDIA、SiliconFlow、ZenMux、智谱、Kimi、MiMo、Mistral 等多家模型 - 自动降级轮询 — 模型失败自动切换，临时错误两轮重试，永久故障 24h 黑名单，全网崩溃 3min 冷却 - CCR 闭环智能体 — 集成工具调用、记忆管理、反思修复、策略进化 - 文件操作 — 读取/写入/列出目录，支持 PDF、Word、Excel 解析 - 联网搜索 — 必应+百度双引擎搜索 - 语音交互 — 支持 edge-tts 语音播报 + speech recognition 语音输入 - 模型管理 — 可视化增删改查、拖拽排序、属地判断 - 多主题 — 深色 / 柔灰 / 浅色三种主题切换 - CCR HTTP 代理 — 兼容 Claude Code 请求格式，自动转发到 OpenClaw 网关 - Self-IDE — 本地 AI Agent IDE 服务 + VS Code 扩展 MyAI Model/ ├── DS/ Streamlit UI 模块 │ ├── main.py 入口文件 │ ├── config.py 配置管理 │ ├── chat engine.py 对话引擎（单次/降级轮询） │ ├── model manager.py 模型增删改查、黑名单、日志 │ ├── tools.py 文件操作、搜索、记忆、插件 │ └── ui.py 完整 UI 渲染（CSS + 组件） ├── CCR/ CCR 闭环智能体 │ └── ccr.py Hermes 智能体 + HTTP 代理 ├── gateway.py FastAPI 网关（多模型路由） ├── openclaw.json 模型池 & Provider 配置 ├── start.bat Windows 一键启动脚本 ├── requirements.txt Python 依赖 └── README.md pip install -r requirements.txt 创建 keys.env 文件，填入你的 API Key： DEEPSEEK API KEY=sk-xxx SILICONFLOW API KEY=sk-xxx NVIDIA API KEY=nvapi-xxx ZENMUX API KEY=sk-xxx ZHIPU API KEY=xxx KIMI API KEY=sk-xxx MIMO API KEY=sk-xxx MISTRAL API KEY=sk-xxx Windows（推荐）： start.bat 手动启动： 启动网关 python gateway.py 启动 WebUI streamlit run DS/main.py --server.port 8502 打开浏览器访问 http://127.0.0.1:8502 { "id": "deepseek:v4-flash-paid", "actual": "deepseek-v4-flash", "provider": "deepseek", "label": "DeepSeek V4 Flash", "type": "auto", "order": 1 } - id — 模型唯一标识 - actual — 实际模型名称（发给 Provider） - provider — Provider 名称（对应 providers 配置） - type — auto （自动降级池）或 manual （手动选择） - order — 优先级排序 { "deepseek": { "baseUrl": "https://api.deepseek.com/v1", "apiKey": "${DEEPSEEK API KEY}", "proxy": "" } } - apiKey 支持环境变量引用 ${VAR NAME} - proxy 为空则使用系统代理 - 前端 : Streamlit + 自定义 CSS - 网关 : FastAPI + Uvicorn - AI : OpenAI SDK 兼容多家 Provider - 语音 : edge-tts / pyttsx3 / SpeechRecognition - 搜索 : 必应 + 百度双引擎 MIT This repo also ships Self-IDE under self-ide/ https://github.com/Alanzj/SelfClaw/blob/main/self-ide/README.md : a local Python service OpenAI-compatible proxy + Goal engine + Subagent orchestration + cost meter, default port 19898 with a thin VS Code extension as its shell. See self-ide/README.md https://github.com/Alanzj/SelfClaw/blob/main/self-ide/README.md for the architecture diagram, HERMES ROOT / GATEWAY PORT configuration and the quickstart.