# Pool types

You choose a pool's type when you create it, under Data Pools → Create data pool. The type cannot be changed afterwards.

# Knowledge (RAG)

For documents: contracts, reports, manuals, presentations, scanned letters, invoices. Documents are split into passages, indexed for search by meaning and by keyword, and described by AI.

When you create one, choose:

• Name and Description. Agents read the description to decide whether to search the pool.
• Access, Private or Organisation. See Access.
• Embedding model, the AI model that indexes the passages for search by meaning. Keep the suggested one unless you have a reason not to; it is set when the pool is created.

Then upload files in Documents.

# Database

For structured data in CSV files: price lists, inventories, transaction exports. Each CSV file becomes a table with typed columns (text, number, true/false, date and time), so an agent can ask for "all rows between two dates" or "the total per region" and get an exact answer rather than a passage of text.

Create the pool with a Name, Access and Description, then import CSV files in Tables.

# External database

For data that already lives in a database. Data Pools connects to it read-only, reads which tables and columns it has, and lets agents query them like CSV tables. Your data stays in your database.

Click Test connection before Create. After the pool is created, its tables are read in the background and appear under Tables.

Good to know:

• The database has to be reachable from the internet. Private and internal network addresses are refused.
• The password is stored encrypted and never shown again.
• For MongoDB, the columns of each collection are worked out from a sample of its documents, and nested objects and lists are left out. Only single-server connections are supported.
• The structure of the database is read when the pool is created. Tables added to the database later do not appear automatically.

# SharePoint (Microsoft 365)

For documents that already live in SharePoint. The pool mirrors the files of one SharePoint site: their names, details and a link back to each original. Files are never copied into Data Pools.

1. Choose SharePoint (Microsoft 365) and click Sign in with Microsoft. Sign in with your own Microsoft 365 account.
2. Pick the SharePoint site. The list shows only the sites your account can open.
3. Leave Index file contents (OCR and RAG) on to make the files' contents searchable. Switched off, the pool only lists the files and their links.
4. Click Create. The first sync runs in the background.

How the sync works:

• The site is synced again every 30 minutes. New files appear, changed files are indexed again, and files deleted in SharePoint disappear from the pool.
• With content indexing on, each supported file is downloaded briefly, indexed like an upload, and not kept. Files over 50 MB, and file types that cannot be indexed, are listed with their link only.
• Agents always get the link to the original file, so they can cite it.
• SharePoint files do not count towards your Data Pool's storage.
• A synced file cannot be deleted in Data Pools. Delete it in SharePoint instead.

---