Lightweight AI agent framework with free providers, a DIY JSON agent loop, and Cloudflare TTS — all served from a single Node.js server.
npm install
cp .env.example .env # fill in your API keys
npm run dev # OpenCode server :4096 + API/UI :3001 + Vite :5173npm run dev also starts opencode serve, which OpenCode Zen's free models need. Without it, run only node server/index.mjs.
- Multi-provider: Groq, OpenRouter, Cloudflare Workers AI, OpenCode Free, Ollama, Mock — all through an OpenAI-compatible interface
- Agent loop: JSON-based reasoning loop with tool calls (weather, calculator, search, code runner, time)
- TTS: Cloudflare Workers AI text-to-speech (English, Spanish, multi-voice)
- Single server: One Node.js process serves both the API and the frontend — no build step needed
- Free to run: Most providers work with no API key or have generous free tiers
| Provider | Auth | Free Tier | Models |
|---|---|---|---|
mock |
none | unlimited | mock-gpt |
opencode |
free: local OpenCode server (OPENCODE_SERVER_PASSWORD) · paid: OPENCODE_API_KEY |
*-free models |
free: mimo-v2.6-flash-free, nemotron-3.5-lightning-free, … · paid: gpt-5.5, claude-sonnet-5, gemini-3.8-flash, kimi-k3, … |
ollama |
none | local | any local model |
groq |
API key | 30 req/min | llama-3.3-70b-versatile, openai/gpt-oss-120b, qwen/qwen3.6-27b |
openrouter |
API key | free models | nvidia/nemotron-3-ultra-550b-a55b:free, google/gemma-4-31b-it:free |
cloudflare |
API key + Account ID | 10K neurons/day | @cf/openai/gpt-oss-120b, @cf/meta/llama-4-scout-17b-16e-instruct, @cf/deepseek-ai/deepseek-r1-distill-qwen-32b |
Cloudflare Workers AI TTS voices:
| Model | Language |
|---|---|
@cf/deepgram/aura-2-en |
English |
@cf/deepgram/aura-2-es |
Spanish |
@cf/deepgram/aura-1 |
English |
@cf/myshell-ai/melotts |
Multi-language |
| Endpoint | Method | Description |
|---|---|---|
/api/providers |
GET | List chat providers and models |
/api/agent/run |
POST | Run agent loop (sync) |
/api/agent/stream |
POST | Run agent loop (SSE stream) |
/api/agent/tools |
GET | List available tools |
/api/tts |
POST | Generate speech audio (MP3) |
/api/tts/providers |
GET | List TTS providers and models |
# OpenCode Zen — free models go through a local `opencode serve`
# (started by `npm run dev`; needs the opencode CLI). Any random string:
OPENCODE_SERVER_PASSWORD=
# Paid Zen models (direct API, needs funds)
OPENCODE_API_KEY=
# Groq
GROQ_API_KEY=
# OpenRouter
OPENROUTER_API_KEY=
# Cloudflare Workers AI
CLOUDFLARE_API_KEY=
CLOUDFLARE_ACCOUNT_ID=
# Ollama (optional, defaults to localhost:11434)
OLLAMA_BASE_URL=omnirouter/
├── server/
│ ├── index.mjs # HTTP server + API routes
│ ├── providers.mjs # Provider registry (reads catalog)
│ ├── agent/
│ │ ├── loop.mjs # Agent loop (reasoning + tool calls)
│ │ ├── tool-registry.mjs # Tool definitions
│ │ └── tools/ # Built-in tools
│ └── free-providers/
│ └── providers-catalog.json # Provider/model catalog
├── frontend/
│ └── public/
│ └── index.html # Single-file frontend (no build)
├── tests/
│ ├── harness.mjs # Test runner
│ └── scenarios.mjs # Test scenarios
├── .env.example
└── package.json
MIT