Skip to main content
Five endpoints, one engine. retrieve and retrieve/stream are the SAME answer from the SAME service: only the way it is delivered differs. drugs and feedback are what comes around it — before a drug card is displayed, after an answer is read.

Which one to call

A script, a batch, a back end

/v2/retrieve. One call, one payload. Nothing to assemble.

A UI where someone is waiting

/v2/retrieve/stream. Steps arrive as they happen, so the wait can be watched.

What you will not find here

The knowledge write plane — PDF ingest, summarization, knowledge-unit extraction — has no route. It runs as a batch, through the locus CLI, where the source PDFs are. An API exposing it would promise work the server cannot do: the PDFs are not on its disk.