tato_code

CLI coding agent that runs against a hosted proxy (see tato-code-backend/). The CLI only needs the proxy URL and a shared token — real LLM and Tavily API keys live on the server.

Setup

# 1. Install
pip install -e .

# 2. Configure (one-time)
cp .env.example .env
# Edit .env and set:
#   TATO_PROXY_URL   = https://your-service.onrender.com
#   TATO_PROXY_TOKEN = the same PROXY_TOKEN you set on the server

# 3. Run
tato

Architecture

┌────────────┐   bearer (TATO_PROXY_TOKEN)   ┌──────────────────────┐
│  tato CLI  │ ───────────────────────────▶ │ tato-code-backend    │
│  (user)    │                               │  (Render, secrets)   │
└────────────┘                               │   ├─ /v1/chat/...    │ ─▶ OpenRouter
                                             │   ├─ /tools/web_...  │ ─▶ Tavily
                                             │   └─ /tools/fetch_.. │ ─▶ Tavily
                                             └──────────────────────┘

The CLI uses langchain_openai.ChatOpenAI pointed at the proxy's /v1 base URL, and the @tool wrappers in src/tato_code/tools/tools.py POST to /tools/web_search and /tools/fetch_url. From the CLI's perspective, OpenRouter and Tavily are just an OpenAI-compatible endpoint and a couple of HTTP routes.

Why a proxy?

Distributing the CLI as a package would otherwise mean shipping the real OPENROUTER_API_KEY and TAVILY_API_KEY to every user. The proxy keeps both keys on the server and gives the CLI a single, token-gated endpoint to talk to.

Configuration reference

See tato-code-backend/README.md for the server-side setup and Render deployment steps.