Meilisearch v1.9 release notes - Meilisearch Documentation

New Features

Hybrid search updates

Meilisearch v1.9 introduces multiple enhancements to hybrid search functionality:

The _vectors field now accepts object values in addition to embedding arrays, allowing you to specify embeddings with additional metadata:

{
  "id": 42,
  "_vectors": {
    "default": [
      0.1,
      0.2
    ],
    "text": {
      "embeddings": [
        [
          0.1,
          0.2,
          0.3
        ],
        [
          0.4,
          0.5,
          0.6
        ]
      ],
      "regenerate": false
    },
    "translation": {
      "embeddings": [
        0.1,
        0.2,
        0.3,
        0.4
      ],
      "regenerate": true
    }
  }
}

The embeddings field replaces a document’s embeddings, while regenerate controls whether embeddings are regenerated on future document updates. Set regenerate: true to import embeddings as a one-shot process, or regenerate: false to preserve embeddings through document updates.

Use the new retrieveVectors search parameter to include the _vectors field in search results (performance improvement makes this opt-in by default):

curl \
 -X POST 'http://localhost:7700/indexes/INDEX_NAME/search' \
 -H 'Content-Type: application/json' \
 --data-binary '{
 "q": "SEARCH QUERY",
 "retrieveVectors": true
 }'

Ranking score threshold

Filter search results by minimum quality using the rankingScoreThreshold parameter:

curl \
 -X POST 'http://localhost:7700/indexes/movies/search' \
 -H 'Content-Type: application/json' \
 --data-binary '{
 "q": "Badman dark returns 1",
 "showRankingScore": true,
 "limit": 5,
 "rankingScoreThreshold": 0.2
 }'

Documents below the threshold are excluded from results and do not count towards estimatedTotalHits or totalHits.

Get similar documents endpoint

Find documents similar to a given document using the new /indexes/{indexUid}/similar endpoint:

curl \
 -X POST 'http://localhost:7700/indexes/:indexUid/similar' \
 -H 'Content-Type: application/json' \
 --data-binary '{
 "id": "23",
 "offset": 0,
 "limit": 2,
 "filter": "release_date > 1521763199",
 "embedder": "default",
 "attributesToRetrieve": ["*"],
 "showRankingScore": false,
 "showRankingScoreDetails": false
 }'

Parameters:

id: Document ID to find similar results for (required)
offset: Number of results to skip (optional, defaults to 0)
limit: Number of results to return (optional, defaults to 20)
filter: Filter expression to apply to results (optional)
embedder: Embedder to use for similarity matching (optional, defaults to "default")
attributesToRetrieve: Fields to include in results (optional, defaults to all)
showRankingScore: Include ranking scores (optional, defaults to false)
showRankingScoreDetails: Include detailed ranking scores (optional, defaults to false)
rankingScoreThreshold: Minimum ranking score threshold (optional)

Supports both GET (URL parameters) and POST (request body) routes.

`frequency` matching strategy

Prioritize results containing less frequent query terms using the new frequency matching strategy:

curl \
 -X POST 'http://localhost:7700/indexes/{index_uid}/search' \
 -H 'Content-Type: application/json' \
 --data-binary '{
 "q": "cheval blanc",
 "matchingStrategy": "frequency"
 }'

Set distinct attribute at search time

Specify the distinct attribute for a search without modifying index settings using the distinct parameter:

curl \
 -X POST 'http://localhost:7700/indexes/{index_uid}/search' \
 -H 'Content-Type: application/json' \
 --data-binary '{
 "q": "kefir le double poney",
 "distinct": "book.isbn"
 }'

A search-time distinct attribute takes precedence over the index settings.

Improvements

Indexing performance

Settings updates are now significantly faster with reduced disk usage. When changing embedding settings, only embedders with modified settings regenerate their embeddings. When only the documentTemplate is modified, embeddings regenerate only for documents where the modification affects the text to embed.

Search performance

Filter AND operations are now faster during search
Facet distribution calculations are optimized for improved performance

Language support

Added new normalizer to normalize œ to oe and æ to ae
Fixed chinese-normalization-pinyin feature flag compilation

Relevancy improvements

All fields now have the same impact on relevancy when searchableAttributes: ["*"]. Fixed searchableAttributes behavior when handling nested fields.

Prometheus metrics (experimental)

Use HTTP path patterns instead of full paths in metrics for better grouping and analysis.

Other

Breaking changes

Empty _vectors.embedder arrays are now interpreted as having no vector embedding (previously interpreted as a single embedding of dimension 0)
The _vectors field is no longer included in search results by default when the experimental vectorStore feature is enabled (use retrieveVectors: true to opt-in)
Meilisearch no longer preserves the exact representation of embeddings in _vectors. Vectors are stored in a canonicalized float representation (e.g., 3 may be represented as 3.0)

Deprecations

The exportPuffinReport experimental feature has been removed. Use logs routes and logs modes instead.

Bug fixes

Fixed security issue in Rustls dependency
Fixed embedding settings reset when changing the source of an embedder, preventing misleading error messages
Fixed panic in hybrid search when removing all embedders
Fixed hybrid search to respect offset and limit parameters when returning keyword results early
Fixed issue where dumps with user-provided embedders and documents opting out of vectors would fail to import correctly (v1.9.1)

Find more information on GitHub

​New Features

​Hybrid search updates

​Ranking score threshold

​Get similar documents endpoint

​frequency matching strategy

​Set distinct attribute at search time

​Improvements

​Indexing performance

​Search performance

​Language support

​Relevancy improvements

​Prometheus metrics (experimental)

​Other

​Breaking changes

​Deprecations

​Bug fixes