Ai Search Endpoint

Turn a plain-language creator brief into a scored, ranked creator table.

A sufficiently specific brief returns a job id at once and queues the search — a deep search reads 600 creators through the model, which can outlast the 120s TimeoutMiddleware. Poll the status route. A plainly broad brief (for example, "give me 10 creators") instead returns status=clarification_required, no job id, and up to three structured questions. Submit the combined answers to this same endpoint with the returned search_id as session_id; the backend retains the original request and starts the job once the brief has a useful subject.

session_id may identify an empty chat created by POST /ai-search/sessions; omit it to create and start a new persistent chat in this call. platforms and size come from the UI's own controls and override whatever the brief implied — a deliberate click beats an inference. depth caps how many candidates the scoring pass reads, which is what keeps cost flat as the corpus grows. live_content=true sources fresh creators from influencers.club and returns them in discovered_creators for the user to save; it is off by default because sourcing spends provider credits. Nothing it finds joins the roster on its own — POST /ai-search/{search_id}/save is what keeps a creator.