Skip to main content
Tags are how you slice a large knowledge base. An Agent may attach to a whole collection, but a tag filter limits it to the subset relevant to that Agent’s job.

​
Two kinds of tags

  • Manual tags — added by you, on upload or later. Full control.
  • Auto tags — added by the Auto-tag enricher in a Foundry pipeline (or in a collection’s ingest defaults). Consistent, scales.
Both write to the same field, both are filterable identically.

​
Tag palette

Your workspace has a shared tag palette — the list of tags anyone can pick from a dropdown. Prevents typo variants (billing, Billing, bills). Add tags to the palette under Workspace settings → Tags. Free-form tagging is still allowed but tags not in the palette are highlighted so you can add them (or fix a typo). Most workspaces end up with tags along a few standard axes:
  • Audience: internal, customer, partner, developer.
  • Domain: billing, auth, pricing, security, support, product.
  • Freshness: current, deprecated, draft.
  • Access: public, restricted, secret.
  • Language: en, ko, ja, es (auto-added by the Language enricher).
Pick a few axes and stick with them. Consistency beats coverage.

​
Folders

Collections support folders — a hierarchical location, separate from tags. A document has one folder and many tags. Folders are useful when the physical layout of your source matters (e.g., mirroring a Google Drive folder tree). Tags are useful when the same document belongs to multiple slices.

​
Filtering by tag in retrieval

Retrieval presets and per-Agent filters take tag predicates:
See Filters.

​
Bulk tag operations

From the collection view:
  • Select multiple documents (checkbox).
  • Add tags or Remove tags in bulk.
  • Or use the CLI: nora knowledge tag <query> --add billing.

​
Auto-tagging from a pipeline

In a Foundry pipeline, add the Auto-tag (LLM) enricher. Give it the list of possible tags and one-line descriptions. Every new chunk gets classified into 0-N tags. Runs incrementally — only new/changed chunks are tagged on subsequent runs.

​
Tag stats

The Insights page shows tag distributions:
  • How many documents / chunks per tag.
  • Coverage — how many docs have zero tags (targets for auto-tagging).
  • Tag frequency in retrieval queries (which tags actually get filtered on).
Useful for spotting unused tags to retire, or gaps in your palette.