AI Sessions now run on OpenAI, Google Gemini, and OpenRouter

AI Sessions are no longer Claude-only. You can now connect your own keys for:

  • OpenAI: plain API key or Azure OpenAI (endpoint + deployment)
  • Google Gemini: AI Studio API key or Vertex
  • OpenRouter: one key gives you their whole model catalog (openai/gpt-5-mini, anthropic/claude-sonnet-4.6, and so on)

Claude keeps working exactly as before, and you can hold credentials for several providers at once.

How to use it

  1. Go to User Settings > AI Terminal and hit Connect on the provider you want. Keys are stored encrypted, per user.
  2. Set the provider (and optionally a model) on a session Profile.
  3. Start a session with that profile. That’s it.

What works on every provider: the full terminal experience, shell and file tools with the same permission and approval rules, MCP servers, budgets (max_budget_usd actually stops the session), hibernation and resume, session forking, and per-provider usage and cost tracking. OpenRouter sessions are metered from OpenRouter’s own billed cost, so any catalog model prices correctly.

Defaults if you don’t pick a model: gpt-5.2 for OpenAI, gemini-3.7-flash for Google, openai/gpt-5-mini for OpenRouter.

If you connected a provider before this went live you may need a refresh for the new Settings rows to appear. Feedback welcome, especially on models you’d like us to tune for.