Getting started

Signing in

Primer does not have its own username and password. It sits behind an identity provider you already run — Google Workspace, Entra ID, Keycloak, anything that speaks OIDC — and trusts the identity that arrives with your request.

For a single-user install, authentication can be switched off entirely. In that mode every request is the same fixed local user, which is exactly right for a laptop and exactly wrong for anything shared. See Configuration.

Your first library

A library is a private collection of sources on one subject. Everything in Primer hangs off a library: documents are uploaded into one, and every question is asked of one.

Start with a narrow library rather than a broad one. A library holding forty papers on a single topic gives sharper answers than one holding four hundred on nine, because retrieval has fewer near-misses to choose between.

Adding sources

Upload the documents you want to be able to ask about. Each upload is processed in the background: parsed, split into passages, and indexed. A document reports its progress while that happens, and becomes searchable when it reaches ready.

Processing takes seconds for a short text file and longer for a large PDF with images, because reading text out of images is slower than reading it out of a text layer.

Asking questions

Ask a question of a library, and Primer retrieves the passages most likely to answer it, then answers using those passages — with a citation for each claim, naming the document, the version, and where in it the passage sits.

If an answer has no citation, treat it with suspicion. The citation is the part you can check, and checking it is the point.

Answers are drafts, not authorities

Retrieval finds passages that look relevant; a model then writes prose from them. Both steps can be wrong. The citations exist so you can verify the parts that matter, and you should.