The Library & RAG

The Library is your curated knowledge base, and RAG is the mechanism that lets a worker ground its answers in your material instead of inventing them.

What you’ll do

1. Add a document to the Library.
2. Ask a worker about it and see it quote your content.
3. Understand how semantic search finds the right material.

1. Add a document

Click 📚 Library in the sidebar, then ⬆ to upload (or drag a file onto the panel). PDFs, .md, .docx, spreadsheets, code, images, audio, and video are all supported. Items live in a tree of kinds — Category, Collection, List, Document, Note, Link — and each can be scoped to a single Company or kept at 🌐 Workspace level for every worker.

2. Ask about it

Open a chat with a worker. When the conversation is scoped to a Company (and always for workspace items), the app retrieves the most relevant Library items and feeds them to the worker as reference. That’s RAG — Retrieval-Augmented Generation. You don’t paste anything; the worker pulls what it needs and can cite it.

Save the good stuff

When a worker produces something worth keeping — a brief, a summary — have it save to the Library. Next time, RAG can surface it automatically.

3. Why it finds the right thing

The search box matches by title or tag for exact lookups. Underneath, the Library also does semantic search using embeddings — so a worker finds material by meaning even when your exact words don’t appear in the document.

There’s a worker for this

The built-in Librarian curates the Library, keeps the RAG index fresh, and files new material so the right document reaches the right worker. Hit 📚 Chat with the Librarian to ask “where’s that doc?”

What to put on the shelf first

An empty Library makes workers generic; a well-stocked one makes them yours. Three high-value additions, in order:

Your two best past deliverables per role. One excellent proposal, one great customer reply, one strong post. This does more for output quality than any amount of prompt tuning, because you’ve stopped describing your standard and started showing it. When a worker drafts something new, it has your actual work to match rather than an average of the internet’s.

The reference material you keep re-explaining. Price list, service descriptions, delivery times, the specifications you get asked about. The test is simple: if you’ve typed it into a chat twice, it belongs in the Library. From then on the worker pulls it and cites it instead of asking you.

The document you’re about to be asked about. Before a meeting, drop in the contract, the brief, or the twenty-page spec, and ask a worker for the five things that matter and anything that contradicts what you were told. This is the fastest useful thing the Library does on day one.

Garbage on the shelf is worse than an empty shelf

RAG retrieves what’s there — it can’t tell that your 2023 price list has been superseded. An outdated document gets quoted with total confidence. Prune as deliberately as you add, and when a price or policy changes, change it in the Library the same day.

Recap

You dropped a document in, a worker quoted it, and semantic search made the match by meaning, not just keywords. That’s RAG: grounded answers from your own shelf. Next, learn how to make a Library doc always present for a worker.

📖 Reference in the docs: Library (screen) · Library (concept)

➡️ Next: Always-on SOPs