Onion

• Extract data from a web page — Submit an extraction and get a job id back immediately. Give a URL and an instruction in plain language ("the product title and price as JSON"). A bare URL with no instruction is valid too, and returns the page itself. Returns in milliseconds; read the outcome with extract_result.
• Extract and wait for the result — Submit an extraction and wait up to wait_seconds for it to finish. Convenience only: on timeout it returns status "processing" and the job id, and the work continues — read it later with extract_result. Pass job_id instead of a prompt to wait on a job that is already running.
• See what the pipeline is doing — The stages a job has been through: planning, crawling, extracting, and the turns of the extraction loop. Use it on a job that is taking a while — it tells "still working" apart from "stuck". Not the outcome; that is extract_result.
• Read an extraction's result — The authoritative outcome for a job id. Note that a successful extraction can legitimately find no data, and that being blocked by the site is a reported outcome rather than a failure.
• Check an extraction's status — Cheap status read for a job id: queued, processing, ready or failed.
• Say whether an extraction was any good — Report how an extraction actually turned out. Call it once you have tried to use a result: you are the only party who can tell that the data was stale, that it answered a different question, or that "no data" was wrong because the page plainly has it — the crawler only knows what it fetched. Any verdict other than "helpful" must say what was wrong. Feedback with no job_id is accepted too, for the service in general.
• Show the current plan — Reports whether this session is anonymous or authenticated, and what the key's plan allows. Call it before using a plan-gated argument such as output_format — or just send one and read the refusal, which names the entitlement that would unlock it.

Schemas and invoke metadata: