Run Pi through AnyRouter

What this is for

Pi ships with its own provider list. Pointing it at a gateway means a custom provider entry, and without one every model switch needs a separate key.

Steps, then copy

1. 1.Install Pi, then create an AnyRouter key in the dashboard (sk-ar-…) and export it as ANYROUTER_API_KEY.
2. 2.Add AnyRouter to ~/.pi/agent/models.json as an openai-completions provider.
3. 3.Open /model in Pi, pick an anyrouter model, and confirm the call in request logs.

curl -fsSL https://pi.dev/install.sh | sh
export ANYROUTER_API_KEY="sk-ar-..."

{
  "$schema": "https://pi.dev/schemas/models.schema.json",
  "providers": {
    "anyrouter": {
      "baseUrl": "https://anyrouter.dev/api/v1",
      "api": "openai-completions",
      "apiKey": "$ANYROUTER_API_KEY",
      "models": [
        { "id": "anyrouter/auto" }
      ]
    }
  }
}

Limits

• Pi reads models.json each time you open /model. Add one entry per model id you want in the picker.

Links

• Pi on GitHub

Next playbooks

• Run OpenCode through AnyRouterOne CLI command injects an AnyRouter provider for the session, or add a custom provider in opencode.json — one key, the Chat Completions catalog, and request logs.
• Run Claude Code through AnyRouterInstall the anyr CLI, then anyr claude — it injects the base URL, key, and model. Env vars are the fallback if you already run claude yourself.
• Run every coding agent through one URLClaude Code, Codex, OpenCode, Cursor, and other OpenAI-compatible tools share 272+ models, one balance, and one request log.