Enrichment & re-indexing
Every knowledge item goes through two stages after you add it: indexing (splitting the content into chunks and embedding them so they're searchable) and enrichment (an AI pass that generates extra metadata to improve retrieval). The item row in Dashboard → Knowledge shows both — a chunk count and an enrichment badge (Enriched, Partial, Pending, or Failed).
Two row actions let you re-run these stages: Reindex and Re-enrich.
Reindex
Reindex rebuilds the item from its content: it deletes the existing embeddings, re-chunks the text, and re-embeds every chunk. Use it when the underlying content changed — you edited the text, the source web page was updated, or chunks look stale or wrong.
Reindex also queues a fresh enrichment pass automatically, so the item's questions and summary regenerate too. In effect, Reindex is the full rebuild.
🖼️ [Image] — The row ⋯ menu showing Edit, Reindex, Re-enrich, and Delete.
Re-enrich
Re-enrich regenerates only the AI metadata — it does not re-embed the passages. Use it when the content is fine but you want fresh enrichment, for example after an earlier pass came back Partial or Failed. It runs in the background and the row's enrichment badge moves to Pending, then updates when done.
Products are enriched by the catalog pipeline, not the knowledge base, so Re-enrich isn't offered for product items.
What enrichment adds
Enrichment runs AI builders over each item to produce metadata that makes search hit more reliably:
- Document summary — a condensed version of the item, embedded so summary-style queries match.
- Hypothetical questions — questions a user might ask that this item answers. These are embedded too, so a visitor's phrasing can match a generated question even when it doesn't match the raw text. The row shows a
…Qcount when present. - Keywords — extracted terms that strengthen keyword search.
Summaries, hypothetical questions, and keywords are all part of the Chat module, so every subscribed workspace gets the full set.
Enrichment is included with the Chat module. A pass can still finish as Partial when the AI couldn't process every part of an item — re-run it with Re-enrich.
Contextual retrieval (shown as a CTX badge) is part of the Chat module too, and is opt-in: it adds surrounding context to each chunk before embedding, for the highest retrieval quality. It's reapplied on Reindex, not on Re-enrich.
Unchanged content is skipped
Indexing tracks a content hash per item. If you trigger a sync and the content hasn't changed, the system skips re-embedding entirely — so republishing a documentation page or re-running a sync won't waste embedding cost on identical text. To force a full rebuild regardless, use Reindex.
🎬 [Video] — Editing an item, hitting Reindex, and watching the enrichment badge cycle Pending → Enriched.