Ask the agent (legacy alias of /v1/ask)

Legacy alias of POST /v1/ask with a minimal response shape. Starts a read-only agent turn (202) to poll via GET /v1/agent/query/{id}, which returns only status and the plain answer text. New integrations should use /v1/ask, which adds structured citations, the verification verdict, and usage. Charges the same flat agent_ask unit.

Bounded synchronous mode (?wait=). Some questions are answered by a cheap deterministic retrieval tier that runs inline, in roughly 1.5-2.5s, rather than by the agent loop. For those, the 202 makes the caller spend a whole poll interval rediscovering an answer the server already has. Send ?wait=<seconds> and an answer produced inside that budget comes back as 200 with the completed payload — the poll body plus turnId. Every other outcome, including a fast-tier answer that overran the budget and every question that goes to the agent loop, still returns 202 { "turnId": ... } exactly as before.

wait is opt-in and additive: a request without it is unchanged. It bounds how long you block before being handed an INLINE answer; it does not cancel work already running, and it never waits on an agent-loop turn. Overrunning the budget costs nothing — the answer is already stored, so your first poll returns it.