Adding documents

What happens to an upload

An upload is checked, stored, and queued before the response comes back. From there, work happens in the background:

A document reports one of these states while you wait:

State	What it means
queued	Stored and waiting for a worker
parsing	Being read and split into passages
chunking	Read; waiting to be embedded
embedding	Passages being turned into vectors
indexing	Written to the index; being checked
ready	Searchable
failed	Something went wrong; see the detail
unsupported	Nothing readable could be found

Duplicate files

If you upload the same bytes twice, Primer stores them once. Both documents exist independently — different names, different libraries if you like — but the file behind them is shared.

This is invisible from the outside, deliberately. Primer never tells you that a file you uploaded already existed, because that would reveal something about content you cannot otherwise see.

Replacing a document

Uploading a replacement adds a new version rather than overwriting the old one. Older versions stay readable, so a citation made last month still resolves to the text that was actually quoted, not to whatever the document says now.

While a replacement is being indexed, the previous version keeps answering questions. The switch happens in one step when the new index is complete and verified — never halfway.

Rebuilding an index

Reindexing rebuilds a document's passages without re-uploading it. Use it after changing the embedding model, or if a document was indexed while something was misconfigured.

The current index keeps answering throughout. The new build only becomes the answer once it is complete and verified, and a build that fails changes nothing.

Pressing reindex twice does not start two builds. A rebuild already in flight is reported as it stands.

Deleting a document

A deleted document stops being searchable at once. Its passages, metadata, and stored file are removed afterwards, in that order.

The order is deliberate. Passages go first because they are what a search can still reach. The file goes last, and only if no other document uses it: if you and a colleague uploaded the same PDF, deleting yours leaves theirs intact and the file in place.

Deleting a document that is already deleted reports that it is not found, the same as any document you cannot see. It does not schedule a second round of cleanup.

Limits

• Uploads are capped by a size limit the operator sets (100 MB by default).
• Very large documents are refused rather than partially indexed, because a document indexed halfway would answer questions from part of its content with nothing to indicate the rest was dropped.