Given an item returned from /discover, run a focused retrieval pipeline pinned to that item’s canonical id and synthesize a 1–2 paragraph brief. Cheaper than a full /search when you only want the high-level picture of a single context point.
Bearer authentication header of the form Bearer <token>, where <token> is your auth token.
Current question. Typically the last user message in the conversation.
1 - 5000One or more tuples of canonical IDs to pin interpretation to. Pass the canonical_ids list from each DiscoveryItem you want to include — tuples are preserved so providers can use the grouping. Example: [["abc", "def"], ["ghi"]].
1Optional prior turns, oldest → newest. Server caps to the last 20 turns.
Storage project id to scope the retrieval against. Omit to span the entire sandbox.
YYYY-MM-DD anchor for time-relative phrasing like 'last week'.
Internal — retrieval preset override. Not part of the public contract yet; defaults to 'auto' when omitted. The 'max' value is reserved and will return 403.
fast, auto, discover, sql, max A macro-level brief grounded in the pinned canonical tuples.
A 1–2 paragraph macro-level summary synthesized from the pinned canonical tuples.
Echo of the items tuples the caller sent, for correlation.
Echo of the sandbox this interpretation ran against.
Echo of the caller's query.
Up to 10 canonical entities that anchored the brief, ordered by relevance.
Echo of the project scope used, or null when the request spanned the full sandbox.