Upload document

PDF and metadata; a new version supersedes the old revision

corpus
7 documents
searchable
73 chunks
Choose
PDF with a text layer
Indexing
parse · chunk · embed
Pending
a person approves it
pdf and metadata12 MB limit · larger sets via the CLI

Larger documents take minutes to embed and belong in the CLI: npm run ingest -- path/to.pdf --title "…"

indexed as pending — approval is what supersedes the revision in force
What happens when this runs.

The PDF is read per page, split along section boundaries, embedded in batches, and written in one transaction. It lands as pending: on file and fully indexed, but invisible to retrieval until a person approves it in Approvals. The revision currently in force keeps answering questions until then, and the supersede happens at approval — atomically, so there is never a moment with two active revisions or none.