Quasar

What it is

Quasar is a research library a company runs for its own people. One catalog holds the papers, the notes, and the files that go with them. A researcher signs in, searches what the organization already has, and can ask two agents. Those agents are only allowed to talk about holdings that person can see.

Who it is for

The main install is an internal cloud for a whole research staff, with company sign-in and collections that not everyone can open. The same application also runs on one machine for a single person or a small lab, if they have the disk. People bring their own model keys. The organization can add shared keys. Quasar does not ship a vendor account, and it is not a public service for strangers.

The catalog

Postgres is the only database. It holds the metadata, the text search, the embeddings, the conversations, and who is allowed to see what. The PDFs live in object storage. Search is lexical and vector together, and the permission check happens inside that query. A result list is never assembled first and filtered afterward. Ingestion workers stay up, so new files keep landing: uploads, a watched folder, or a source an admin turned on. Quasar does not crawl the open web.

The agents

A librarian files, tags, finds duplicates, and answers what the catalog already contains. It cannot invent a paper that is not there. A research assistant writes briefs, comparisons, and reading lists, and it has to retrieve a passage before it says what a paper claims. Every citation is a chunk the caller is allowed to read. Voice uses that same librarian. It is not a second character.

Kepler

Quasar does not include a reader. Opening a work hands it to Kepler as that signed-in person, and notes can come back into the catalog.

Where the work stands

The contract for that system is written. There is no application in the repository yet.