Skip to main content

Create a search index

This is the step-by-step for creating an index, configuring its fields, and loading data. About 10 minutes.

Before you start

  • You're logged in as an admin or moderator.
  • If you want semantic or hybrid search, you have an AI provider configured.
  • You have some data in JSON, JSONL, CSV, or Excel format. If you don't, the Initial Setup demo gives you a ready-made Fashion Catalog.

1. Open the create wizard

Sidebar → Capabilities → Search IndexesNew Index.

2. Fill in basic info

FieldWhat to put
Display nameHuman-readable. "Product Catalog", "Help Articles".
Index nameAuto-generated from the display name. Lowercase + hyphens. Can't be changed later.
DescriptionOptional. "Main product index used by the storefront search."
Search typePick Hybrid unless you have a specific reason. See Search indexes → The three types of search. Can't be changed later.
Search providerElasticsearch (default). Pick Azure AI Search only if you've configured that. Can't be changed later.

Click Next.

3. Search settings

FieldWhat to put
Indexing strategyOn Upload for most cases. Documents become searchable as you upload them.
LanguageThe primary language of your content. Drives stemming and stop-words.
SynonymsSkip for now — see Synonyms and stop words once you've seen the index in action.
Custom stop wordsSkip for now.
Number of replicas (ES)0 is fine for dev.
Refresh interval (ES)Default 1 second is fine.

Click Next.

4. AI configuration (Semantic / Hybrid only)

FieldWhat to put
AI providerDefaults to your system default. Override if needed.
Embedding modelDefaults to your provider's default. Can't be changed later.
Vector similarity metricCosine (default). Don't change without a reason.

(For Hybrid only) RRF rank constant and window size — leave at defaults. They tune how lexical and semantic results blend, and the global defaults under Settings → Search are usually right.

Click Create Index.

5. Configure fields

You land on the index detail page with no fields and no data. Click Configure Mappings (or Fields in the header).

  1. Paste a sample of your data (10 records is enough) in the Source Data panel on the right.
  2. Interakt detects the fields and proposes types. A dialog asks you to confirm — review the list.
  3. Click Create Fields.

The fields appear in the table. For each one, click the gear icon to open the configuration sidebar:

  • Searchable — turn on for fields users search by words (title, description, brand).
  • Facetable — turn on for fields users filter by (category, brand, colour, size, status).
  • Sortable — turn on for fields you'll let users sort by (price, date, rating).
  • Include in response — usually yes; off for internal-only fields.
  • Boost value — leave at 1 except for the 1–2 most important fields (title often gets 2).

For full coverage of each switch, see Index fields.

Click Save Mappings in the floating save bar at the bottom.

6. Upload data

Click Upload in the index detail page header.

  • Drag your file into the drop zone or click to browse.
  • Leave batch size at 1000.
  • Click Upload.

A progress bar shows. For OpenAI, expect ~30 seconds per 1000 documents. Ollama is several times slower.

The index status badge turns green ("Ready") and the document count updates when it's done.

7. Verify

Sidebar → Playground → Index Search. Pick your index. Try a few queries:

  • A literal term from your data — should be the top result.
  • A vaguer phrase that doesn't appear word-for-word — with hybrid, semantic matches should fill the gaps.
  • Apply a filter on a facetable field — confirm filtering works.

If the results look right, you're ready to wrap this index in a search experience.

Troubleshooting

Upload failed: "could not generate embeddings." Your AI provider isn't reachable. Check Platform → AI Providers → Test Connection. If Ollama: confirm ollama serve is running and the embedding model is pulled.

Some documents failed to upload. The Recent Uploads panel on the upload page shows which batch failed and why — usually a single malformed record. Fix it and re-upload that batch.

Field types look wrong after upload. Edit the field on the Fields page. Some changes need a rebuild to apply to existing documents.

"Index name already exists." Names are unique system-wide. Delete the existing one or pick a different name.

Where to go next