For the complete documentation index, see llms.txt. This page is also available as Markdown.
Search
A search is one literature query over the corpus, run at a chosen depth. It produces an Answer and the ranked papers that answer is grounded in.
Lifecycle
Async. POST /v1/searches returns 202 with status: queued; poll, subscribe to a
webhook, or use the SDK's .wait(). Terminal states: succeeded, failed, canceled.
Shape
{ "object": "search", "id": "srch_9dm2pq4x1a", "status": "succeeded", "depth": "standard",
"query": "How does climate change affect biodiversity?", "answer": { "…": "…" },
"paper_count": 20, "credits_cost": 0, "created_at": "…", "completed_at": "…" }Depth changes the shape of the work
standard scans fewer papers and returns faster; deep_review runs an agentic pass across more
papers and takes materially longer. Cost and latency both scale. See depth.
Not to be confused with
Chat — search finds papers and answers across the corpus; chat answers questions about a fixed set of documents you name. If the caller already knows which PDFs matter, they want a chat. Topic — search returns papers; topics return the ideas those papers are about.
Related
Last updated