Pi supports every provider listed in the coding agent README. With the corresponding credentials (API key or OAuth login) you can use:
API key providers
- Anthropic
- OpenAI (plus Azure OpenAI with deployment mapping)
- Google Gemini (API key mode)
- Mistral
- Groq
- Cerebras
- xAI
- OpenRouter
- Vercel AI Gateway
- ZAI
- OpenCode Zen
- MiniMax (global and China tenant)
- Amazon Bedrock (via AWS profile, IAM keys, or Bedrock bearer token)
OAuth providers (via /login)
- Anthropic (Claude Pro/Max accounts)
- GitHub Copilot (GPT‑4o, Claude, Gemini through Copilot)
- Google Gemini CLI (Gemini 2.0/2.5)
- Google Antigravity (Gemini 3, Claude, GPT‑OSS sandbox models)
- OpenAI Codex (ChatGPT Plus/Pro)
Beyond the built-ins you can register any OpenAI-compatible or Anthropic-compatible endpoint through ~/.pi/agent/models.json, so local runtimes or proxies (Ollama, LM Studio, vLLM, etc.) work too. Availability is gated only by which keys you provide.
