Managing Collections

A collection is a group of documents you query together — a course, a project, a client, a research topic. In the app they appear in the sidebar under Knowledge Stacks.

Collections are the main lever you have over answer quality. Retrieval searches what you point it at, so a focused collection gives focused answers, and a single collection holding everything you own gives vague ones.

Creating a collection

1. Open Knowledge Stacks in the sidebar
2. Click New Knowledge Stack
3. Give it a name
4. Optionally add a description — or click Generate with AI to have one written from the contents
5. Optionally choose a parent collection to nest it; leave it empty for a top-level collection
6. Click Create

The description is not decoration: it is what automatic collection routing reads when it decides which collections a question belongs to. A collection described as "lecture notes and papers on Bayesian inference" gets picked for the right questions; one described as "stuff" does not.

Adding documents

1. Select the collection
2. Drag files onto the Upload tab, or click Browse Files
3. Click Compose once the files you want are staged
4. Watch live progress as each document is parsed, chunked and embedded

Documents become queryable as soon as processing finishes. You can also fill a collection from a connector (Google Drive, Notion, Zotero and others) or by importing papers by DOI. See Uploading Documents for formats and processing detail.

How many collections you can have depends on your plan: 3 on the free Researcher plan, 50 per workspace on Pro, unlimited on Team and Enterprise.

Nesting

A collection can have a parent, which gives you a hierarchy:

Doctorate
├── Methodology
│   ├── Qualitative
│   └── Statistics
└── Literature review

Nest when a subject genuinely splits into parts you would want to query separately. Do not nest for the sake of tidiness — every extra level is another decision you have to make when asking a question.

Querying a collection

Automatic — ask your question and let Scrapalot route it. It reads your collection names and descriptions, picks the ones that look relevant, and tells you which it chose.

Explicit — scope the question yourself from the chat box:

Type	Result
@	Pick an individual file
@/	Pick a collection
@/name/	Browse inside a named collection

Pinned — pin a collection in the Knowledge Stacks popover while AI Routing is active, and routing searches only that collection, overriding auto-selection. Useful when you know exactly where the answer lives and want no interference from the rest of your library.

Sharing

Collections belong to a workspace. Everyone in a workspace sees its collections, so sharing a body of work means putting it in a shared workspace rather than sharing collections one by one. See Collaboration.

Housekeeping

• Rename or edit — open the collection and choose Edit Knowledge Stack
• Delete a document — remove it from the collection; its embeddings go with it
• Delete a collection — removes the collection and its documents. There is no undo, so export anything you want to keep first.
• Audio overview — generate a 10–15 minute podcast-style conversation between two hosts about the collection

How to organize well

Split by what you would ask about, not by where files came from. "Downloads" is not a collection; "Immunology — reading list" is.

Keep collections small enough to be specific. A hundred closely related documents work well together. A thousand unrelated ones dilute every search.

Write real descriptions. They drive automatic routing, and they are what your collaborators read.

Separate reference material from working material. A collection you query constantly and an archive you rarely touch do not belong together.

Next steps

• Uploading Documents — formats, OCR, processing
• Asking Questions — search modes and Deep Research
• Collaboration — workspaces and shared notes