Library

The Library is askTheodor’s knowledge base: a tidy, tagged, folder-like store of your material — documents, notes, lists, links, and uploaded files. It’s the shelf your Workers reach for when they need facts that live in your world, not on the open web.

The Library is what makes workers answer from your reality instead of guessing. That technique has a name: RAG.

What RAG means (plainly)

RAG stands for Retrieval-Augmented Generation. It’s a simple idea: before a worker answers, the app retrieves the most relevant pieces from your Library and feeds them to the model alongside your question. The model’s reply is then grounded in your actual documents — fewer made-up answers, more “according to your notes.” You curate the shelf; the app fetches the right book at the right moment.

How it’s organized

The Library is a tree — items can sit inside other items, like folders and files. Each item has a kind:

• Category — a folder, pure structure.
• Collection — a curated grouping of related material.
• List — an ordered list (markdown bullets).
• Document — long-form writing.
• Note — a short jotting.
• Link — a saved URL, shown as a clickable card.

(Uploaded files — images, spreadsheets, and the like — live here too.) Items can be tagged for quick filtering, and a sub-tree can be scoped to one Company or kept global for the whole workspace.

Three kinds of memory, kept separate

It helps to know what the Library is not:

• The Library is your deliberately curated knowledge — you decide what’s on the shelf.
• Per-worker memories are what an individual worker quietly learns and recalls during its own runs.
• Per-conversation work products are the outputs a worker produces inside a single chat.

Keeping them distinct means your trusted reference material never gets muddled with a worker’s scratch notes.

Save the good stuff

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

A worker just for this

There’s a built-in Librarian who specializes in curating the Library, keeping the RAG index fresh, and filing new material so the right document reaches the right worker.

See the Library screen to browse, add, and organize your knowledge.

In practice

Scope decides the audience — choose it deliberately. Workspace scope reaches every worker in every Company; Company scope stays with that team. Client material, per-project context and anything confidential belongs at Company scope. Reserve Workspace for things that are true everywhere: your own brand voice, your policies, your standards.

Stale documents are the main failure mode. RAG retrieves what’s on the shelf and quotes it with complete confidence — it has no way to know a price list was superseded in March. When a price, policy or delivery time changes, update the Library the same day. A wrong answer delivered confidently is worse than no answer.

Prune as deliberately as you add. Several near-duplicate versions of the same document make retrieval worse, not better: the worker may ground its answer in whichever one matched semantically, which won’t reliably be the newest. Keep one current version and delete the rest.

High-value first uploads, in order of payback: your two best past deliverables per role (they show the standard rather than describing it), the reference material you keep re-explaining, and whatever document you’re about to be asked questions about.

The “typed it twice” test

If you’ve pasted the same context into a chat twice, it belongs in the Library. That’s the moment the upload pays for itself.

🎓 Learn it hands-on: The Library & RAG

Terms in this page

• Library — your curated, tagged, tree-structured knowledge base of documents, notes, lists, links, and files.
• RAG (Retrieval-Augmented Generation) — the technique of fetching relevant Library material and feeding it to the model so its answers are grounded in your content.
• Grounded — describing an answer that’s based on supplied source material rather than the model’s general guesswork.
• Tree / parent–child — the folders-within-folders structure of the Library; an item can live inside another.
• Kind — what a Library item is: category, collection, list, document, note, link (or an uploaded file type).
• Tag — a label you attach to items for filtering and search.
• Scope (company / global) — whether a Library item belongs to one Company or to the whole workspace.
• Worker / persona — the AI character that reads from the Library; see Workers.
• Memory — what a single worker learns and recalls during its runs; distinct from the Library.
• Work product — output a worker creates inside one conversation; also distinct from the Library.