Rebuilding an index

Sometimes you change something and existing documents need to be reprocessed for the change to apply. This is "rebuilding" or "reindexing." It's a single button click — the screen tells you when you need to use it.

When you need to rebuild

A rebuild reprocesses every document in the index against your current field config and text-analysis settings. You need it when any of these have changed since the documents were uploaded:

What changed	Why it needs a rebuild
Language	The text analyser used to tokenise documents at index time is set by the language. If you change it, existing tokens are stale.
Synonyms	Synonyms are baked in during indexing. New synonyms only apply to new uploads until you rebuild.
Stop words	Same — applied during indexing.
A field's type (text → keyword, number → keyword, etc.)	The way the value is stored physically changes.
Searchable toggled on a field	Searchable fields are tokenised on the way in; non-searchable ones aren't.
Facetable toggled on a field	Facetable fields are stored as exact tokens for filtering and aggregation.
Autocomplete toggled on a field	Autocomplete uses a special analyser that needs to be applied at index time.
Filter value mappings changed	The canonical-value normalisation happens at index time.
Field renamed	The new name doesn't exist in existing documents until they're reprocessed.

Things that do not need a rebuild:

• Boost values — used at search time, not index time. New boosts apply immediately.
• Display name of a field — purely cosmetic.
• Include in response — applies at query time.
• Description of the index.
• Number of replicas (Elasticsearch).
• Indexing strategy (changing how future uploads are batched).

How the UI tells you a rebuild is needed

Two visual cues:

• On the Edit screen, changing language / synonyms / stop words shows an amber banner: "Reindexing required. These changes will require a full reindex of all documents to take effect."
• On the Fields screen, changing field attributes that affect indexing shows a "Reindex needed" badge at the top, and the save bar at the bottom says "Changes require reindexing after save."

Saving doesn't trigger the rebuild — you save first, then click Reindex when you're ready.

How to rebuild

From the index detail page

1. Open the index.
2. Look at the Index Name card on the right.
3. Click Reindex.
4. A dialog asks to confirm and shows the current document count.
5. Click Reindex.
6. The status badge changes to Indexing and a progress bar shows.
7. When done, the badge goes back to Ready and the success dialog shows stats (documents processed, duration).

From the Fields screen

After changes that require a rebuild, the floating save bar has a Reindex Now button. Same flow.

What happens during a rebuild

• Every document is re-read from internal storage.
• Run through the current text-analysis pipeline (language, stop words, synonyms).
• Re-mapped through your current field mappings.
• Written back to the search index.

For Semantic / Hybrid indexes, the embeddings are not regenerated — only the lexical side is reprocessed. This is fast.

To regenerate embeddings (e.g. because you changed which fields feed into the embedding), you need to re-upload the documents. The rebuild button doesn't do this — see Loading data.

How long does it take

Index size	Typical rebuild time
Up to 1,000 documents	Seconds.
1,000 – 50,000 documents	Tens of seconds to a couple of minutes.
50,000 – 1,000,000 documents	Minutes to ~half an hour.
Larger	Plan a window — run during low-traffic time.

A rebuild is non-blocking — search keeps working against the existing index while the rebuild runs. When it completes, the new tokens take over.

When something goes wrong

If a rebuild fails, the index status goes to Error and a recovery banner appears on the detail page with a Recreate Index button.

Recreate Index is the heavy-hammer fallback. It:

1. Drops the underlying index structure in the search provider.
2. Rebuilds the empty structure from your saved field definitions.
3. Leaves you with an empty index — you re-upload your documents.

This is destructive of indexed data but not of configuration. Your field mappings, synonyms, language, AI config, and so on all survive. Use it when:

• The search provider got into a bad state ("shards corrupted", "mapping conflict") that a reindex can't fix.
• You changed something that's only allowed on a fresh index (rare).

Switching the AI provider or embedding model

This is a different and more destructive operation, not the same as a reindex.

The Edit screen → AI Config tab has a Change Configuration button. Clicking it warns you:

Switching to a different provider or model will delete all indexed documents and require you to re-upload your data to generate new embeddings.

This is because different embedding models produce different-sized vectors and Interakt's vector storage is hardcoded to one size per index. There's no in-place migration — you reset, re-upload, and the new model embeds everything from scratch.

If you really need to migrate to a different embedding model, the safest path is:

1. Create a new index with the new model.
2. Upload your data to it.
3. Test the new index alongside the old one in the Playground.
4. Once you're happy, point your search experience at the new index (and deactivate the old one).

This is also the right pattern for any change that needs a rebuild but you want zero downtime — build the new one parallel, switch, retire the old one.

Rebuild vs re-upload — which one do I need?

You changed…	You need…
Synonyms, stop words, language	Reindex
Field attributes (searchable, facetable, autocomplete)	Reindex
Field types	Reindex
Filter value mappings	Reindex
The AI provider or embedding model	Reset + re-upload (via Change Configuration)
Which fields feed into the embedding	Re-upload
Boost values	Nothing — applies on next query
Display configuration in the search experience	Nothing — applies on next query
Display name / description / "include in response"	Nothing — applies on next query

Common gotchas

• Forgetting to click Reindex. You save the change, the UI says "saved", and you wonder why searches still behave the old way. Look for the "Reindex needed" badge.
• Rebuilding doesn't regenerate embeddings. If the change is about which fields embed (or which model), you have to re-upload.
• The status badge says Indexing for a while. That's fine — the rebuild is running. Search keeps working in the meantime.
• Recreating an index loses your data. Use only when reindex can't fix the problem.

Where to go next

• Index fields — what's behind the toggles that need a rebuild.
• Synonyms and stop words — the most common reason to rebuild.
• Loading data — what to do after Recreate Index.