Last 7 days
0
Features: 0
Changes: 0
Fixes: 0
Deprecations: 0
Meilisearch
Search-as-a-service platform powered by the Meilisearch engine.
Latest Meilisearch changelog updates, official release notes, breaking changes, security patches, pricing changes, and developer reactions in one product feed.
Follow this Meilisearch release-notes page to spot useful features, risky migrations, noisy announcements, and source links before they hit your backlog.
Where Meilisearch updates are checked
Changes.Watch links back to official changelog and release-note sources so summaries stay easy to verify.
Related changelog channels
Use channels to follow groups of tools around a stack, workflow, or topic.
Meilisearch updates by type
Rolling windows show how many product updates landed in the last 7, 30, 90, and 365 days, grouped by existing changelog semantics.
0
7
20
64
Latest Meilisearch changelog updates
Added environment variable to limit concurrent LMDB read transactions for the task queue.
Reverts a recent search speed optimisation introduced in PR #6542 to fix related issues
Reverted the `/tasks/stream` and `/batches/stream` Server-Sent Events (SSE) endpoints.
Make the health route checks blocking
Added sharding support for foreign filters, increasing document retrieval limit to 1,000 and allowing network‑based evaluation
Added experimental SSE streaming endpoints `/tasks/stream` and `/batches/stream` for real-time task and batch updates.
Added filter‑condition support and a `lastUpdatedAt` timestamp to Dynamic Search Rules, plus new experimental DSR fuel environment variables.
Breaking changes to Dynamic Search Rules API (request types, mandatory selector IDs, async task responses) and default network behavior for federated document fetch in sharded setups
Improved synonyms storage with lazy loading, delivering up to 13× query performance gains.
Fix race condition in S3 snapshot uploads that could abort the snapshot.
Fixed CVE‑2026‑57823 and CVE‑2026‑57824 vulnerabilities that could cause privilege escalation and information disclosure
Patched CVE-2026-57824 preventing privilege escalation through overly permissive API keys.
Revert commit 6432 to address a dumpless upgrade bug
Add experimental POST /render route to render templates or fragments, gated by the renderRoute feature flag.
Added search personalization support for federated search and made the new settings indexer feature‑complete, boosting performance and scaling.
Add experimental `queueDocumentsFetch` feature that queues GET /indexes/:uid/documents and POST /indexes/:uid/documents/fetch when no processing thread is available.
Fixed regression from v1.45.0 that incorrectly batched deletions by filter with document additions/updates.
Fix internal vector store quantization config desync
Reverts unintended deletions caused by autobatch filter operations.
Introduced a new settings indexer that speeds up indexing of many settings (displayed fields, synonyms, primary key, etc.) and can be disabled via an env var or CLI flag.
Remote federated facet search is now the default for network sharding, with a new useNetwork flag and a breaking change tying embedder timeout to the index’s searchCutOffMs.
Fix for an authenticated SSRF vulnerability in Meilisearch
Introduced a new settings indexer with filterable, sortable, facet search and custom attribute support, optional for Cloud and toggleable in OSS via env var.
Fixed legacy settings indexer bugs and resolved internal errors when removing or modifying fragments with the experimental multimodal feature
Added remote fallback for sharding/replication with /network status exposure
Introduced Dynamic Search Rules (experimental) for condition‑based pinning of documents via PATCH/DELETE endpoints.
Added support for a global distinct attribute in federated search, enabling cross‑index distinct computation and facet distribution (query‑level distinct now disallowed).
Add experimental `foreignKeys` index setting to hydrate documents from other indexes (cross‑index document hydration).
Fix regression where Meilisearch could stop processing tasks after an automatic task queue cleanup in v1.38.0–v1.38.1
Fix regression that caused the engine to stop processing tasks.
Improved embedding indexing speed by adopting the latest hANNoy and eliminating full database scans
Introduces replicated sharding with a new `network.shards` field, changing the PATCH /network contract and requiring automatic migration of existing networks;
Added new ranking criteria `attributeRank` and `wordPosition` plus a vector store migration to Hannoy, improving relevance and ranking scores.
Fixed task database corruption when using S3 streaming snapshots during snapshotting.
Added `showPerformanceDetails` flag to search‑related endpoints, returning `performanceDetails` for better observability.
Increased the auth store database size to 2 GiB, enabling indexing of more keys.
Fixed regression from v1.34.1 where Meilisearch failed to start when the configuration file omitted experimental allowed IP networks.
Fixes a security vulnerability that allowed Meilisearch to send requests to private IPs, now blocking non‑global IP addresses by default.
Added optional `useNetwork` flag to search endpoints (POST, GET, multi‑search) enabling federated network search with pagination support.
Security patch fixing a path‑traversal vulnerability in dump import (no action needed on Cloud, update recommended for open‑source users)
Added new POST /indexes/{indexUid}/fields endpoint to provide detailed field metadata for indexes.
Prevent panic on dumpless upgrade of empty indexes with configured embeddings
Add MEILI_EXPERIMENTAL_DISABLE_FID_BASED_DATABASES_CLEANUP env var to skip field ID‑based database cleanup during upgrades from <1.32.0
Added detailed progress tracking and logging for search operations to improve observability.
S3 streaming snapshots are now Enterprise‑only, removing them from the Community Edition (breaking change) and requiring a license.
Fix task attribution during index swap to prevent cross-index task loss.
Adds ability to change the number of participants in automatic sharding via a new leader field and version UUID, enabling dynamic scaling without re‑sending all data.
- New settings indexer now supports searchableAttributes, exactAttributes, proximityPrecision and embedders, with progress display and instant cancellations; disabled by default for Cloud and toggleable via env var.
Fixes high memory usage and OOM crashes on /metrics route when many tasks exist by limiting retrieved tasks to one
Fixed uploading of Linux amd64 binaries for both Community and Enterprise editions.
Added separate Enterprise (BUSL‑1.1) and Community (MIT) binaries and Docker images, with the Enterprise binary distinguished by an “enterprise” suffix.
Improved task ingestion by defaulting batch size to half of max indexing memory, boosting performance.
Added `customMetadata` query parameter to document‑related task routes, exposing metadata in task responses and webhooks.
Introduced experimental search personalization using Cohere reranking and support for uploading snapshot tarballs directly to S3.
Improved vector store search cutoff handling for experimental indexes
Added a compact index route that queues a defragmentation task, boosting search and indexing speed up to 4×
Makes remote federated search timeout configurable via the MEILI_EXPERIMENTAL_REMOTE_SEARCH_TIMEOUT_SECONDS environment variable
Increased remote federated search timeout from 5 seconds to 30 seconds
Update zerometry to latest version with support for collection, lines, and multi‑line features
Introduced a new geo backend with a filterable geojson field and support for geoPolygon, geoBoundingBox, and geoPoints filters.
Added experimental vector store backend via new vectorStoreSetting feature for better performance with binary quantization.
Added progressTrace display for in‑progress batches.
Optimized hybrid search with filters to greatly reduce query latency
Added automatic document sharding for horizontal scaling (Enterprise only).
Added queryVector field in search responses when retrieveVectors is enabled.
Added publishing of OpenAPI spec as a release asset
Added chat completions route and native webhook API support; "STARTS WITH" filter now stable without experimental flag
Introduces experimental multimodal embeddings, enabling indexing and searching of images alongside text with support for image query parameters.
Fixed a bug where searchable fields removed from filterableAttributes became unsearchable despite being listed in searchableAttributes.
Introduces experimental conversational and LLM‑driven chat features for Meilisearch indexes
Added typoTolerance.disableOnNumbers to disable typo tolerance for numeric queries
Added granular filterable attribute settings to enable/disable specific filter types per attribute for indexing performance optimization.
- Updated the embedded mini dashboard to version 0.2.18.
Added support for dumpless upgrades across all v1.13 patches
Fixed typo tolerance bug and corrected dumpless upgrade logging
AI‑powered search is now enabled by default; breaking changes include removal of the vectorStore experimental value, stricter Ollama URL formats, and renamed error codes for embedder and hybrid queries.
Added support for configuring the max readers limit via an environment variable.
Fix stuck indexation caused by internal BBQueue capacity overflow
Facet strings are now unified by their normalized value, improving consistency across the API.
Fix dump export failure when the task queue includes enqueued or processing tasks
Fix infinite loop bug in indexing process under certain conditions.
Fixed incorrect field distribution calculation for incremental document updates via PUT route
Fix performance regression when updating a small number of documents in indexes with sortableAttributes or filterableAttributes.
Fixed engine bug where adding an empty payload caused batch failures
Indexing speed dramatically improved (up to 4× faster) with new facetSearch and prefixSearch settings to skip parts of the indexing process.
- Added retry logic for REST/OpenAI/Ollama autoembedders when remote response deserialization fails
Introduce timeout settings for read operations.
Add a 3‑second timeout to embedding requests made during search
Experimental AI‑powered search improvements: hybrid.embedder becomes mandatory, new embedder options (binaryQuantized, documentTemplateMaxBytes) and default model change, with binaryQuantized offering huge indexing speed and size gains b...
Introduces experimental CLI flags to configure search concurrency per core (nb searches per core).
Reactivates the Swedish tokenization pipeline when the index language is set in settings
Improved search handling under heavy load by consolidating to a single queue, ensuring scheduler continuity, releasing permits promptly, and capping processing time.
Fixed an issue where embedding an empty document list caused errors, now returning an empty list of embeddings.
Introduces federated search with merged result lists and index weight configuration
Adds support for generating vector embeddings in dumps, enabling smoother migrations.
Added AI‑powered similar documents endpoint, rankingScoreThreshold, frequency matching strategy, distinct search parameter and multiple hybrid search enhancements.
Fix memory leak in the core engine
Fix concurrency bug causing race conditions in the search engine
Fix indexing of geo fields when modifying settings after documents are present.
Added hybrid search enhancements: new Ollama and generic REST embedder sources, a distribution embedder setting, and support for negative keyword operators to exclude terms.
Update grenad library to resolve rare database corruption
Patched a security vulnerability in the HTTP/2 implementation that could be exploited to slow down instances.
Fixed regression from v1.7.1 where adding sortable and filterable nested attributes failed to trigger reindexing
Bump version to v1.7.3
Added a timeout to webhook requests to prevent hanging calls.
Skip reindexing when unknown faceted fields are modified, improving indexing speed
Added experimental AI‑powered search with vector store, GPU‑accelerated HuggingFace embeddings, and new OpenAI embedding models (text‑embedding‑3‑small/large) with configurable dimensions.
Disable incremental facet update as a temporary workaround for facet-related bugs
Restore highlighting for hybrid search when possible
Added experimental automated embeddings (OpenAI, HuggingFace, user‑provided) and hybrid search combining keyword and semantic results.
Fixed search on exact attributes using attributeToSearchOn (issue #4233).
Improved indexing speed for text‑heavy datasets (5‑20% faster) and added Khmer language support.
Fix regression search bug with distinct attribute when using page/hitsPerPage (issue #4140).
Improved indexing speed for dump imports by using buffered readers/writers.
Added nonSeparatorTokens, separatorTokens, and dictionary index settings for customizable word segmentation and domain‑specific dictionaries.
Reverted the v1.3.3 filter escaping fix that forced backslashes to be escaped
Fixed statistics for document deletions performed via filter.
Fix stats calculation for document deletions performed via filter
Fixed filter escaping issues causing errors (4038)
Temporarily disable an assert in the bucket sort algorithm to prevent a panic in ranking rules.
Fixed handling of geo bounding box queries when coordinates are provided as strings; existing documents must be re‑indexed.
Experimental vector search added, allowing storage of embeddings and nearest‑vector queries via new /experimental features endpoint.
Introduced new filter operators IS NULL and IS EMPTY plus routes to delete and fetch documents by filter.
Upgrade to v1.1.1 resolves issue 3664 and disables auto‑batching of additions/deletions.
- Added Multi-Search endpoint to query multiple indexes in one request and removed limits on index count and size.
Fixed panic when searching with stop words in quotes.
Fixed pagination issue when using the page parameter in search requests.
Added Korean support and enhanced Chinese processing plus broader diacritics normalization for several languages
Upgrade to v0.30.5 as it contains critical fixes and should replace earlier 0.30.x releases
Fix pagination issue referenced in issues 3258 and MeiliSearch/milli #743
Replace v0.30.2 with v0.30.3, addressing critical bugs
Fix filter bugs (issues 727 and 3217)
v0.30.1 replaces v0.30.0 and warns that persisting bugs indicate a corrupted database.
Replaces v0.29.2; if bugs persist your database is corrupted and you must reindex the data.
Add zero‑downtime index deployment via a new swap‑indexes API and support for a configuration file (config.toml).
Fixed duplicate‑document bug in the search endpoint
Fixed glibc compatibility by making Meilisearch work with libc 2.27 (Debian 10).
Added matchingStrategy "all" to require all query terms in search results.
Update mini dashboard to version 0.2.1.
Added new features: smart‑crop algorithm, deterministic API‑key creation with UID, Hebrew language support, and improved tokenizer contribution flow;
Introduce experimental auto‑batching feature (use v0.27.2 instead of v0.27.1)
Fix import dumps bug (issue 2396)
Add full nested field support with dot notation and customizable typo‑tolerance settings, plus new Japanese language tokenizer
Resolved a panic crash caused by the auto-batching functionality
Introduced tenant tokens (JWT‑based per‑user access) enabling multi‑tenant indexes
Fixed filtering when using camelCased attributes
Fix task duration calculation
Introduce API key management with new default Search and Admin keys, replace public/private keys, add /keys endpoint, and switch to Authorization: Bearer header (feature, security, breaking).
Breaking: error handler fields renamed and error codes/types updated, introducing new structured error objects and many new error types.
Breaking release fixing critical primary‑key inference bug; requires automatic re‑indexing and enforces Content‑Type headers on all payload routes.
Critical bug in primary‑key inference makes this version deprecated; users must upgrade to v0.23.1 and re‑index all documents, with mandatory Content‑Type headers for payload routes (breaking change).
Introduce a new indexer with automatic re‑indexing, fixing crashes, RAM usage and improving indexing speed (breaking change)
Fix compilation panic when .git directory is missing
- Breaking changes: rename attributesForFaceting to filterableAttributes, replace filters/facetsFilter with filter (only “=” syntax supported, “:” syntax deprecated), rename MAX MDB SIZE env var and fieldsDistribution, update /version re...
/health endpoint now returns 200 instead of 204 (breaking change)
Added integration with the new tokenizer.
Fixed corrupted data handling in placeholder searches and corrected HTTP maintenance errors.
Add automatic index creation when pushing documents to a non‑existent index