Skip to main content
A Knowledge Base is a collection of documents — supplier datasheets, brand guidelines, prior catalogs, domain references, policy files — indexed for retrieval and cited in operation outputs. When an operation draws from a Knowledge Base, it tells you which document and passage it used, so every generated value is traceable. Knowledge Bases are workspace-scoped (shared across catalogues) but attached per-operation (you choose which KBs a given run can see).

​
How they work

  1. Upload documents (PDFs today; Word, PowerPoint, and plain text coming soon).
  2. Claro chunks and indexes the content automatically.
  3. When an operation runs, the model retrieves the most relevant passages per record.
  4. Outputs include citations — document name, page number, the relevant passage.
Retrieval is semantic, not keyword-based. A product description about “wireless over-ear headphones with active noise cancellation” will pull the relevant section of a supplier datasheet even if the wording differs.

​
When to use a Knowledge Base

​
Storage limits by plan

*Higher quotas available on request.

​
Setting up a Knowledge Base

Step 1 · Create
  • From the Knowledge Bases panel, click New Knowledge Base.
  • Use a descriptive name: Brand-Guidelines-EN, Supplier-Datasheets-2025, REACH-Compliance-Docs.
Step 2 · Add content
  • Drag and drop PDFs, or paste text directly.
  • Organize by topic — focused collections retrieve better than broad dumps.
  • Update documents when the source changes; old versions can be archived rather than deleted so existing provenance links stay valid.
Step 3 · Attach to an operation
  • When configuring Bulk Enrichment, Generative Engine, or a Research Agent run, select the relevant Knowledge Bases.
  • Multiple KBs can be selected; the model searches across all of them.

​
Best practices

  • Stay focused. One KB per topic is more accurate than a single large collection. The retriever scores passages against the query; fewer irrelevant passages means better hits.
  • Keep current. Archive old versions; don’t leave outdated docs alongside current ones.
  • Use descriptive names. Include purpose and date so it’s clear what each KB contains and whether it’s current.
  • Test relevance. After attaching a KB, check the citations on a sample of outputs to confirm the right passages are being retrieved.
  • Multilingual data. Create separate KBs per language rather than mixing locales in one collection.

​
Deleting a Knowledge Base

Deleting a KB does not affect existing operation results — they retain their citations and provenance. It only means future runs can no longer access that content.