🔑 AI tokens & models

Your squad runs on your AI accounts - ChatGPT (Codex), Claude, Claude Code, Grok, Gemini, GitHub Copilot, Kimi, MiniMax, Z.AI, Qwen Cloud, and OpenRouter - connected in Settings → AI Tokens. No markup, no middle-man metering: model usage bills to your subscription like your own usage does.

Multiple accounts per provider are first-class: the numbered priority order IS the fallback chain - if account one hits a limit, the runtime tries the next. Reorder with the arrows, save, done. The sync banner tells you when credentials haven’t reached every agent (one click fixes it), and expiry shows before it bites - with a Reconnect right there.

Models per job: workspace defaults set the standard model; individual agents and individual schedules can override - run the heavy reasoning on your best model and the hourly checks on a fast one. Every run’s detail shows which model actually ran.

Fallback models: Agent Configuration also takes an ordered list of fallback models. If your primary is ever unavailable (expired connection, rate limit, an outage), agents automatically try the next model in the chain instead of stalling. The list never offers your current primary as its own fallback, so the chain always moves to a genuinely different model.

Effort and time, also defaults: Agent Configuration carries a workspace reasoning effort default (how hard models think before answering - raise it for deep work, lower it for quick checks) and a default timeout for agent runs, so a stuck run ends instead of hanging forever. Both apply everywhere unless a schedule overrides them.

💡 Tip: If agents suddenly stop, check here first: the expiry banner at the top of the dashboard almost always knows why. Reconnect takes under a minute.