@co-browse/proxy

A small Fastify server that sits between @co-browse/react and your LLM provider. It streams responses over SSE and keeps provider API keys off the client. It speaks a neutral message + tool-call format and translates it per provider, so the same client works across Anthropic (Claude), OpenAI, Groq, Together, Fireworks, OpenRouter, DeepSeek, xAI, Mistral, Ollama, or any OpenAI-compatible endpoint.

Run as a server

npx @co-browse/proxy

Configure via env (see the repo's .env.example):

PROVIDER=anthropic          # or openai | groq | together | … | openai-compatible
MODEL=claude-opus-4-8       # provider default used when omitted
ANTHROPIC_API_KEY=...       # the key env matching PROVIDER
ALLOWED_ORIGINS=http://localhost:5173
# Optional hardening: AUTH_TOKEN, RATE_LIMIT, MAX_BODY_BYTES

Embed in your own Fastify app

import { buildServer } from "@co-browse/proxy";

const app = await buildServer({
  config: { provider: "anthropic", apiKey: process.env.ANTHROPIC_API_KEY!, model: "claude-opus-4-8" },
  allowedOrigins: ["https://your-app.com"],
});
await app.listen({ port: 8787 });

Full env/provider reference: https://github.com/originalkamaal/co-browser-ai#readme

License

MIT © Kamaal Khan