POST — ask a question
Response:
What happens during a turn
The agent answers with the latest briefing, your memory facts, and semantically-matched past episodes in context. On top of that it has a live tool belt (up to 5 tool steps per turn):- Keyword ranks & suggestions, competitor lookups, ASO score
- App Store Connect metrics, reviews, subscription data
- Apple Search Ads and RevenueCat summaries
- Your pending action queue — including approve/reject by id when you explicitly ask
GET — list past Q&A
Cost
The model comes from your LLM settings (provider + BYOK key, or platform defaults). Typical cost per turn is 0.005 depending on model and how many tools the agent used.Scope and limits
- Always scoped to your
owner_email— the agent can never see another user’s data. - The agent can approve/reject existing proposed actions, but cannot originate new write actions from chat — those come from the regular run pipeline with the usual review flow.
- For a full re-analysis (fresh collector pass), trigger a run instead.
- For “why did metric X move?” investigations with ranked root causes, use the Why-agent.

