What meilisearch/meilisearch shipped
Written by FoxPlug from the repository's own releases; not affiliated with Meilisearch. Paste your own repository in the box below and FoxPlug writes the same weekly update for your product. This page is an automatic summary of the releases, pull requests and commits in github.com/meilisearch/meilisearch; Meilisearch did not write it and does not use or endorse FoxPlug. Every line links to the change it describes.
Get a weekly update like this for your product, free
Week of November 2, 2026
What shipped
- Meilisearch v1.55.1 was released with important security fixes, and upgrading from v1.55.0 and earlier is recommended for all users. Release
- Meilisearch v1.56.1 was released with important security fixes, and updating from v1.56.0 is recommended for all users. Release
- V1.56.1 fixed an issue with ranking scores greater than 1. Release
- V1.56.1 made sure all nested fields are inserted in the field id map. Release
- A pull request addressed outdated CA certificates in the Docker images by bumping the base image. Pull request #6705
- A pull request backported some fixes for release in v1.55. Pull request #6707
- The version in Cargo.toml was updated for the next release, v1.56.1. Pull request #6709
- The latest stable release of Meilisearch was published. Release
Why it matters
Two patch releases this week carry security fixes, and the maintainers recommend that everyone on earlier versions upgrade. The releases also correct ranking scores above 1 and ensure nested fields are fully registered in the field id map, so search results and filtering on nested fields behave correctly.
Changelog entry
- V1.55.1: important security fixes; upgrading from v1.55.0 and earlier is recommended. Release
- V1.56.1: important security fixes; updating from v1.56.0 is recommended. Release
- Fixed issue with ranking scores greater than 1. Release
- Made sure all nested fields are inserted in field id map. Release
- Fixed outdated CA certificates in the Docker images. Pull request #6705
- Backported some fixes for release in v1.55. Pull request #6707
- Updated version for the next release (v1.56.1) in Cargo.toml. Pull request #6709
- Published the latest stable release of Meilisearch. Release
Meilisearch v1.55.1 and v1.56.1 are out with important security fixes. Upgrading is recommended for all users. These releases also fix ranking scores greater than 1 and make sure all nested fields are inserted in the field id map.
Meilisearch v1.55.1 and v1.56.1 are out, both with important security fixes. Upgrading from v1.55.0 and earlier, and updating from v1.56.0, is recommended for all users. The v1.56.1 release also fixes an issue with ranking scores greater than 1 and makes sure that all nested fields are inserted in the field id map. Separately, a pull request addressed outdated CA certificates in the Docker images, and fixes were backported for release in v1.55.
Week of October 26, 2026
What shipped
- Released v1.56.0, adding a sampledCount sorting option for facet distribution that gives a rough estimate rather than an exact count of documents per facet. Release
- Fixed ranking scores that could exceed u32 when intermediate ranks were multiplied, causing wraparound and wrong final scores, by storing ranks as u64. Pull request #6703
- Fixed the runtime and memory consumption of long filters containing many tokens, and adjusted filter error messages for multicolumn spans and multi-line filters. Pull request #6695
- Fixed a panic when modifying the searchable attributes list after a document was updated to insert new nested fields, ensuring all nested fields are inserted in the field id map. Pull request #6700
- Clamped
map_sizeto page size when removing tasks during snapshot creation, since heed rejects amap_sizethat isn't a multiple of the page size. Pull request #6699 - Fixed Debian CI, which had been failing since v1.54.0. Pull request #6698
- Reduced cargo flaky iterations to 60 to prevent the job from timing out. Pull request #6689
- Updated the version for the next release to v1.56.0 in Cargo.toml. Pull request #6697
- Published prototype-v1.55.0-revert-threshold-behavior.0, reverting a merge from the fix-ranking-score branch. Release
Why it matters
This week brought a new release with a sampledCount option for faster, approximate facet distributions, plus fixes for ranking scores that could wraparound, long filters that consumed too much memory and time, and a panic when changing searchable attributes after nested field updates. Users on affected versions get more reliable scoring and filtering and can recover from the nested field issue by adding a document with the new fields.
Changelog entry
- V1.56.0: added sampledCount sorting option for facet distribution, a rough estimate that is less precise than counting documents per facet. Release
- Fixed ranking scores greater than 1 by storing ranks as u64 instead of u32 to avoid overflow and wraparound. Pull request #6703
- Fixed memory consumption and runtime of long filters with many tokens; adjusted filter error messages for multicolumn spans and multi-line filters. Pull request #6695
- Fixed panic when modifying searchable attributes after inserting new nested fields; all nested fields are now inserted in the field id map. Pull request #6700
- Clamped
map_sizeto page size when removing tasks during snapshot creation. Pull request #6699 - Fixed Debian CI failing since v1.54.0. Pull request #6698
- Reduced cargo flaky iterations to 60 to prevent job timeouts. Pull request #6689
- Updated version for the next release to v1.56.0 in Cargo.toml. Pull request #6697
- Published prototype-v1.55.0-revert-threshold-behavior.0, reverting a merge from the ranking score fix branch. Release
V1.56.0 is out with sampledCount, a faster approximate option for facet distribution. Also fixed: ranking score overflow on large intermediate ranks, high memory use on long filters, and a panic when changing searchable attributes after nested field updates.
Meilisearch v1.56.0 is out. It adds sampledCount, a new sorting option for facet distribution that trades precision for a rough estimate and is less expensive than counting. This week also fixed ranking scores that could wrap around when intermediate ranks grew large, reduced the memory consumption and runtime of long filters with many tokens, and fixed a panic when changing searchable attributes after a document was updated with new nested fields; existing databases hitting that issue can be repaired by adding a document containing the new nested fields. Snapshot creation and Debian CI also got fixes, and the next release version was bumped in Cargo.toml.
Week of October 19, 2026
What shipped
- Added a sampledCount sorting option for facet distribution that gives a rough estimate of facet order while keeping per facet document counts exhaustive. Pull request #6649
- Updated the code to use the latest version of heed for a safer and more correct way of opening databases, bumping heed, cellulite, arroy and hannoy. Pull request #6687
- Fixed the empty request body section in the settings API reference, which affected all 22 settings routes. Pull request #6678
- Updated the README for clarity, adding runnable code and coverage of common questions. Pull request #6692
- Published a prototype release building on the better database opening work and addressed clippy warnings. Release
Why it matters
This week brings a new facet distribution sorting option that trades precision for speed on large result sets, plus a database opening update aimed at correctness. The settings API reference now renders request bodies for all settings routes, so developers can see the expected payloads. The README was also updated to explain more clearly what the project is and how to get started.
Changelog entry
- Add sampledCount sorting option for facet distribution, giving a rough facet order while keeping per facet document counts exhaustive. Pull request #6649
- Use the latest version of heed for safer, more correct database opening; bump heed, cellulite, arroy and hannoy. Pull request #6687
- Fix empty request body section in the settings API reference across all 22 settings routes. Pull request #6678
- Update README for clarity with runnable code and answers to common questions. Pull request #6692
- Release prototype build better-db-opening and make clippy happy. Release
This week: a new sampledCount option for facet distribution sorting gives a faster rough estimate while keeping per-facet counts exact. Also: safer database opening with the latest heed, and all 22 settings routes now show request bodies in the API reference.
This week in Meilisearch: facet distribution can now be sorted with sampledCount, which returns the facets more quickly with a rough estimate of their order, while the document count for each facet stays exhaustive. We also updated to the latest heed for a safer and more correct way of opening databases, and fixed the settings API reference so all 22 settings routes display their request body. The README was updated for clarity, with runnable code and answers to common questions about the project.
Week of October 12, 2026
What shipped
- Released v1.55.0, which adds an optional inlineRule parameter to search query objects so a dynamic search rule can be previewed inline on a search query. Release
- Fixed a bug where an embedder created after deleting a binary quantized embedder could be wrongly treated as binary quantized by hannoy. Pull request #6679
- Indexing now gives up when an embedder cannot handle the load, with an adaptive rate limiting strategy that accepts slow indexing for small usage with few documents. Pull request #6681
- Optimized the next function in state by replacing the set expression with an intersection-based one, improving performance about four times. Pull request #6571
- Updated the ranking score threshold documentation. Pull request #6676
- Published the latest stable release of Meilisearch. Release
- Updated CONTRIBUTING.md and the PR template to state that generative AI tooling is disallowed in external contributions. Pull request #6684
Why it matters
This week brings a new stable release with an inline way to preview a dynamic search rule before applying it. It also fixes an embedder bug that could mislabel a new embedder as binary quantized after an earlier one is deleted, and changes indexing behavior so it stops when an embedder cannot keep up. Contributions policy is also tightened regarding generated tooling.
Changelog entry
- Add an optional inlineRule parameter to search query objects so a dynamic search rule can be previewed inline on a search query. Release
- Fix wrong binary quantization state after deleting embedders, where a later non-quantized embedder was treated as binary quantized by hannoy. Pull request #6679
- Give up indexing when an embedder can't handle the load, with an adaptive rate limiting strategy. Pull request #6681
- Optimize the next function in state, improving performance about four times. Pull request #6571
- Update the ranking score threshold documentation. Pull request #6676
- Disallow generative AI tooling in external contributions in CONTRIBUTING.md and the PR template. Pull request #6684
- Latest stable release of Meilisearch. Release
Meilisearch v1.55.0 is out: preview a dynamic search rule inline with the new inlineRule parameter on search queries. Also this week: a fix for binary quantized state after deleting embedders, and indexing now gives up when an embedder can't handle the load.
Meilisearch v1.55.0 is out. Search query objects now accept an optional inlineRule parameter so you can preview the effect of a dynamic search rule on a query before applying it. This week also fixes a bug where an embedder created after deleting a binary quantized embedder could be wrongly treated as binary quantized, and indexing now gives up when an embedder can't handle the load, using an adaptive rate limiting strategy. The ranking score threshold documentation was updated, and the contributing guide and PR template now state that generative AI tooling is disallowed in external contributions.
Week of October 5, 2026
What shipped
- Added an optional inlineRule parameter to search query objects in several endpoints, allowing an inline dynamic search rule to be applied directly in a search query. Pull request #6669
- Published a prototype release that avoids panics during highlighting by using get rather than bracket indexing and recovering when the value is absent. Release
- Prevented a panic during search highlighting with crop when the segmenter returns byte offsets inside a unicode character. Pull request #6673
- Updated the version for the next release to v1.55.0 in Cargo.toml and Cargo.lock. Pull request #6675
- Prevented dequantizing by passing binaryQuantized as null. Pull request #6656
- Made various allocation changes, including preventing a concurrent dbi open in heed and switching from mimalloc to jemalloc. Pull request #6654
- Fixed ranking score threshold performance. Pull request #6661
- Changed document retrieval to replace a dynamic check with types, removing unreachable code paths. Pull request #6670
- Published a prototype release that supports the edge case where four or more rules exist in the dynamic search rules index. Release
- Fixed flaky tests by changing the runner image to one that provides cargo and by marking a very long test as ignored to reduce CI time. Pull request #6674
Why it matters
This week brings a preview of dynamic search rules as an inline search parameter, plus a prototype release that makes highlighting with crop robust against incorrect segmentation and a fix that prevents a panic there. Search results should also be faster thanks to the ranking score threshold performance fix, and the dequantizing reset issue is addressed. Release preparation for v1.55.0 is underway.
Changelog entry
- Add optional inlineRule parameter to search query objects in several endpoints for preview of dynamic search rules. Pull request #6669
- Prototype release avoiding panics in highlighting by using get instead of bracket indexing. Release
- Prevent panic during highlighting with crop when the segmenter returns byte offsets inside a unicode character. Pull request #6673
- Update version to v1.55.0 in Cargo.toml and Cargo.lock. Pull request #6675
- Prevent dequantizing by passing binaryQuantized as null. Pull request #6656
- Prevent concurrent dbi open in heed and switch from mimalloc to jemalloc. Pull request #6654
- Fix ranking score threshold performance. Pull request #6661
- Change document retrieval to replace dynamic check with types, removing unreachable code paths. Pull request #6670
- Prototype release supporting the edge case where four or more rules exist in the dynamic search rules index. Release
- Fix flaky tests by changing runner image and marking a long test as ignored. Pull request #6674
- Fix OpenAPI docs by removing duplicated Render template route and unused docs points. Pull request #6667
- Add GitHub Actions workflow to enforce benchmarks label on each pull request. Pull request #6663
This week: preview of inline dynamic search rules, a prototype release avoiding highlight crop panics, faster ranking score threshold search, prevention of dequantizing by passing binaryQuantized as null, and release prep for v1.55.0.
This week in Meilisearch: a preview adds an optional inlineRule parameter so an inline dynamic search rule can be applied to a search query. A prototype release fixes a panic during highlighting with crop when the segmenter returns byte offsets inside a unicode character, by using get rather than bracket indexing and recovering when the value is absent. Ranking score threshold performance is fixed, and dequantizing is prevented by passing binaryQuantized as null. Document retrieval now replaces a dynamic check with types, removing unreachable code paths, and a prototype DSR release supports the edge case where four or more rules exist in the index. Version v1.55.0 is prepared in Cargo.toml and Cargo.lock.
Week of September 28, 2026
What shipped
- V1.52.4 released with a stability fix addressing crash issues on thread exit. Release
- V1.53.3 released with a stability fix addressing crash issues on thread exit. Release
- V1.54.3 released with a stability fix addressing crash issues on thread exit. Release
- V1.54.2 released fixing task queue issues for users of the dynamic search rules experimental feature when upgrading from v1.53 or prior. Release
- PR #6658 replaces mimalloc with jemalloc and patches heed to remove crash issues on thread exit in v1.54. Pull request #6658
- PR #6659 replaces mimalloc with jemalloc and patches heed to remove crash issues on thread exit in v1.53. Pull request #6659
- PR #6660 replaces mimalloc with jemalloc and patches heed to remove crash issues on thread exit in v1.52. Pull request #6660
- PR #6651 fixes dynamic search rule update tasks created in Meilisearch versions prior to v1.54 from causing internal errors when processed, displayed, deleted or canceled. Pull request #6651
- PR #6644 adds facet distribution progress steps to the search progress trace. Pull request #6644
- PR #6650 adds a link to the MCP guide in the generated API reference. Pull request #6650
Why it matters
Three patch releases address a critical stability issue across multiple version lines. The underlying fix replaces the memory allocator and patches database transaction handling to eliminate thread exit crashes that affected users across v1.52, v1.53, and v1.54 branches.
Changelog entry
- V1.52.4: Stability fix for thread exit crashes Release
- V1.53.3: Stability fix for thread exit crashes Release
- V1.54.3: Stability fix for thread exit crashes Release
- V1.54.2: Fixed dynamic search rules task queue issues on upgrade from v1.53 or prior Release
- V1.54 edition: Replaced mimalloc with jemalloc and patched heed for unique transactions Pull request #6658
- V1.53 edition: Replaced mimalloc with jemalloc and patched heed for unique transactions Pull request #6659
- V1.52 edition: Replaced mimalloc with jemalloc and patched heed for unique transactions Pull request #6660
- Dynamic search rule update tasks from v1.53 and prior now process correctly in v1.54 Pull request #6651
- Added facet distribution progress steps to search progress trace Pull request #6644
- Added MCP guide link in generated API reference Pull request #6650
Meilisearch v1.52.4, v1.53.3, and v1.54.3 are now available with important stability fixes. Update recommended.
We've released stability updates across three version lines: v1.52.4, v1.53.3, and v1.54.3. These releases address thread exit crash issues by replacing mimalloc with jemalloc and introducing database transaction improvements. Additionally, v1.54.2 fixes task queue handling for dynamic search rules when upgrading from earlier versions. Update recommended for all users.
Week of September 21, 2026
What shipped
Changelog entry
- Breaking: v1.54.0 updates dynamicSearchRules experimental features API Release
- Fix embedder setting changes when API keys are expired Pull request #6640
- Ignore geo errors from existing documents when updating or deleting Pull request #6638
- Aggregate analytics from the MCP route Pull request #6639
- Replace mimalloc with jemalloc for memory management Release
- Update heed dependency and remove
non_free_page_sizeRelease - Use experimental heed branch with unique transactions Release
- Detect and remove missing and corrupted tasks from task queue Release
Meilisearch v1.54.0 is out with breaking changes to dynamicSearchRules. Fixed embedder API key expiration handling and geo error resilience in document operations.
Meilisearch v1.54.0 is now available. This release includes breaking changes to the dynamicSearchRules experimental features, with automatic migration for Cloud users. We've also fixed embedder setting changes when API keys expire and improved resilience by making geo errors non-fatal during document updates and deletions.
Week of September 14, 2026
What shipped
- The POST /dynamic-search-rules endpoint now returns actions as an object with pin and other fields instead of a list, a breaking change to the dynamic search rules API. Pull request #6549
- An experimental MCP route was added supporting multi-index search, facet search, index descriptions, and index name listing. Pull request #6629
- A verbosity mode was introduced to the Progress trace that allows creating quiet progress traces that do not register logs when not in use. Pull request #6631
- Grammar and broken Markdown links in error code descriptions were fixed. Pull request #6626
- Documentation fixes corrected a mailto typo in the README, a broken TOC anchor in CONTRIBUTING.md, and grammar issues. Pull request #6619
- Release prototype-v1.52.3-fix-crashes-with-debug.2 added mimalloc tracking with callbacks and callback registration. Release
- Release prototype-v1.52.3-fix-crashes-with-debug.1 added symbols in the release build. Release
Why it matters
This week includes a breaking change to the dynamic search rules API alongside experimental features for MCP integration and improved tracing capabilities. Bug fixes to documentation and internal improvements to crash handling support ongoing stability.
Changelog entry
- Breaking: POST /dynamic-search-rules endpoint now returns actions as an object with pin and other fields instead of a list Pull request #6549
- Added experimental MCP route supporting multi-index search, facet search, index descriptions, and index name listing Pull request #6629
- Introduced verbosity level to progress trace for quieter logging in tests and documentation Pull request #6631
- Fixed grammar and broken Markdown links in error code descriptions Pull request #6626
- Fixed documentation typos and broken links in README and CONTRIBUTING files Pull request #6619
- Improved debug builds with mimalloc callback tracking Release
New this week: experimental MCP route for multi-index search, verbosity mode for progress traces, and the dynamic-search-rules endpoint restructured. See the details on what shipped.
This week's updates to Meilisearch include an experimental MCP route for multi-index search and facet operations, a new verbosity mode for progress traces that quiets logs in tests, and a breaking change restructuring the dynamic-search-rules endpoint actions field. Documentation fixes and crash handling improvements round out the week.