Question 1

What is librocat?

Accepted Answer

librocat is an open-source knowledge base shared by all your AI agents and everyone on your team. Tell one agent something once, and every agent knows it. Libro, an AI librarian, keeps it organized and up to date. It's free to start at librocat.com, and free to self-host.

Question 2

Which agents does it work with?

Accepted Answer

Coding and desktop agents: any agent that supports Agent Plugins, or MCP with Agent Skills (not web chat). Claude Code, Codex, Cursor, and GitHub Copilot each install it with one or two commands, shown at the top of this page. To add it to every other agent on your computer at once, run npx plugins add laughing-hermit/librocat if they support Agent Plugins, or use MCP and Skills; both are on the Connect page. Desktop agents install the same plugin: Claude Desktop and Cowork today, the ChatGPT desktop app through a workspace admin (or just its MCP without one), and Grok Bot through a Cursor Teams admin. Each agent signs in once in your browser, where you pick its library. There are no API keys.

Question 3

Can my whole team share one library?

Accepted Answer

Yes, on the Team plan. Each teammate connects their own agent to the same library, in any mix: two of you on Claude Code and one on Cursor, or everyone on the same agent. Whatever one agent saves, every teammate's agent can find. Everyone signs in as themselves, as an Admin, an Editor, or a Reader.

Question 4

How is this different from agent memory or a vector database?

Accepted Answer

An agent's memory belongs to that one agent, and a vector database returns text that only looks similar. librocat is shared by all your agents and your whole team, and it's organized like a real library, with a catalog of titles, descriptions, and tags, a thesaurus, links, review dates, and a librarian. Search matches the words you asked for, in any of their forms, so you can see why something came up and whether it's still true.

Question 5

What does Libro do?

Accepted Answer

Libro is librocat's librarian (and a cat). When an agent saves a concept without a description or tags, Libro fills them in. It reads every file you add from start to finish, describes it, and flags the concepts it likely changes. Each week it flags duplicates and answers past their review date (on Free, in the weeks your agents write), suggests what to replace, retire, or give a review date, and checks the merges your agents draft. You approve or dismiss each one in Needs you, and every change is saved as a new version.

Question 6

Is my data private?

Accepted Answer

Yes. We don't sell your data, use it for ads, or train AI models on it. Libro's model, from TypeSafe AI, reads your concepts and the files you add in order to catalog them, with zero data retention: nothing it reads is kept after it answers. Search uses no model at all. The privacy page lists exactly what Libro reads. You can export or delete everything anytime, or run librocat on your own server.

Question 7

What can I add to it?

Accepted Answer

Anything your agents write down, plus documents (PDF, Word, spreadsheets, slides, text, and code) and pictures, up to 50 MB a file on Free and 200 MB on paid plans. On a paid plan you can also save a link: librocat keeps a copy of the page, or of the file it points to, up to 50 MB (a PDF, an Office file, or a picture). Some sites, like LinkedIn, don't let apps read their pages. Agents find everything by its words, and pictures by their titles. For a recording, save its transcript.

Question 8

What are the limits on each plan?

Accepted Answer

Free: 1,000 concepts, 100 MB of files (10 MB of it text, 50 MB a file), 200 of Libro's credits, 10,000 calls, 1,500 writes, and 1 GB of file transfer a month. Individual: 5,000 concepts, 5 GB of files (250 MB of text, 200 MB a file), 500 credits, then $0.01 each if you turn on pay-as-you-go, up to a monthly cap you set, 30,000 calls, 5,000 writes, and 10 GB of file transfer a month. Team: 15,000 concepts, 25 GB of files (1 GB of text, 200 MB a file), 2,500 credits, then the same, and 30,000 calls and 5,000 writes a month for each person, with 50 GB of file transfer a month, shared by up to 10 libraries you own, on one bill, where each person is an Admin, an Editor, or a Reader (their agents only read). A credit is a concept Libro catalogs, or about three pages of a file you add; files use at most half. Saving links is on paid plans. Self-hosted librocat has no limits.

Question 9

What happens if I cancel?

Accepted Answer

You keep your library. It goes back to Free with everything in it, and new concepts and files wait until it's under Free's limits. On a Team, your teammates pause right away. Free has one library, so 30 days later your others pause: agents can't reach them, but you can still open, export, or delete them. Choose a plan again and everything comes back. If something is still over Free's limits after 12 months, it's deleted, and we email you twice before that happens.

Question 10

Is it open source? Can I host it myself?

Accepted Answer

Yes, all of it, under Apache-2.0, including the code that runs librocat.com. Run it with Docker on your own server, free and with no limits, and move a library between the two with Export and Import.

The open-source knowledge base for AI agents.

Three steps. Under a minute.

1. Install it

2. Tell it once

3. Every agent knows

Everything your agents learn, in one library you own.

Every agent, every teammate

Kept current by Libro

Your documents, findable

Yours to keep

Pricing

Questions

Stop repeating yourself.

	Free	Individual	Team
Libraries	1	1 per plan	Up to 10, sharing these
Concepts	1,000	5,000	15,000
Files (of it, text)	100 MB (10 MB)	5 GB (250 MB)	25 GB (1 GB)
Largest file	50 MB	200 MB	200 MB
Save a link	No	Yes	Yes
Libro's credits a month (a concept cataloged, or ~3 pages of a file you add)	200	500, then $0.01 each if you turn it on	2,500, then $0.01 each if you turn it on
Calls a month	10,000	30,000	30,000 a person
Writes a month	1,500	5,000	5,000 a person
File transfer a month	1 GB	10 GB	50 GB