Chat, smart routing, and planned moves
The personal assistant talks when you chat, looks things up when you ask, and plans multi-step work only when needed.
Chat, smart routing, and planned moves
Your personal assistant no longer treats every message as a tool call.
- Everyday conversation gets a normal reply (with what it remembers).
- Simple look-ups (calendar, balance, inbox) run one read and summarize.
- Complex work gets a plan of next moves first, then thinks and acts one step at a time — with confirm before risky changes.
Each turn picks a model for the job (fast for chat, stronger for hard agent work). You can steer that with optimizeFor on AI settings: cost, balance, or intelligence.
Call POST /v1/api/agi/runs/{id}/step with mode: "auto" (default), or force chat / agent. Responses include phase, routing, and when relevant reply or agentPlan.
See the API story Chapter 23 and Choose models and bring your own key when that guide exists.
