Delightful search built on object storage

1. Run node

docker run -d --name exofind -p 8080:8080 \
  -v exofind-data:/data \
  -e EXOFIND_AUTH_ROOT_KEY=exok_quickstart \
  ghcr.io/levelfourab/exofind:main-latest

2. Define index

PUT /v1alpha1/admin/indexes/books
Authorization: Bearer exok_quickstart
Content-Type: application/json

{
  "fields": {
    "id": { "type": "string", "primaryKey": true, "required": true },
    "title": { "type": "string", "matching": { "typoTolerance": {} }, "sort": {} },
    "category": { "type": "string", "filter": {}, "facet": {} },
    "published": { "type": "boolean", "filter": {} }
  }
}

3. Search

POST /v1alpha1/indexes/books/search
Authorization: Bearer exok_quickstart
Content-Type: application/json

{
  "query": [
    { "type": "text", "text": "silent spr" },
    { "field": "published", "match": { "value": true } }
  ],
  "limit": 20
}

What it does

Object storage is the source of truth

Wipe a node and it pulls the index back from the bucket. Nodes coordinate with conditional writes, without Raft or cluster coordinators.

Search powered by Lucene

Full-text matching with typo tolerance, prefix autocomplete, geospatial filters, nested sub-documents, and k-nearest neighbors (KNN) vector search.

Per-value language analysis

Documents carry their language tag. The engine applies CJK segmentation, stemming, and Scandinavian compound splitting automatically without custom pipeline setup.

Declarative schemas and index generations

Send the schema you want with a PUT request. When a change requires a reindex, Exofind builds a new generation in the background and promotes it while queries continue.

Shared keys without a database

API keys are stored in the bucket. A key created on one node propagates to all nodes within seconds, scoped by permissions and index patterns.

Interchangeable stateless nodes

Read nodes never talk to each other. Nodes answer every query from local disk and scale horizontally behind standard load balancers.

See it searching

Each demo searches a real dataset through a node, and exists to show one thing the engine can do.

• Swedish food search

  Open demo

  2 606 foods from Livsmedelsverket, the Swedish Food Agency. Shows compounding splitting of words, type sås and the results include gravlaxsås.

  • compound splitting
  • facet counts that ignore their own filter
  • range buckets
  • highlighting

• Airport type-ahead

  Open demo

  Every airport with an IATA code, from OurAirports. Type sto and Stockholm appears at once. Type the typo stockhlm and Stockholm is still there.

  • typo tolerance
  • prefix matchers
  • per-field weights
  • distance filters and sorts

• Cleveland Museum of Art

  Open demo

  30 000 objects from the open access collection with a rich ability to filter them.

  • facet counts that ignore their own filter
  • range buckets
  • ordering
  • numbered pages and cursors
  • highlighting

• Record shop

  Open demo

  300 000 records from Discogs, and every pressing of them.

  • sub-documents
  • facets and ordering inside a value
  • matched values
  • prices and values read out of the search box

Where to go

Tutorials

Learn the engine by running it.

• Getting started
• Getting started with object storage

How-to guides

Solve one task at a time.

• Defining an index
• Indexing documents
• Tuning ranking

Reference

Look up a setting, a request or an error code.

• Search API
• Field types
• Configuration

Explanation

Find out why the design is the way it is.

• Synchronization
• What a write guarantees
• Relevance

Who is behind this

Exofind is built by Level Four AB. The engine is open source under the Apache License 2.0, and you can run it without us.

We also work with the teams that use it: index design, deployment on your own object storage, and tuning relevance. Tell us what you are building at hello@levelfour.se.