Manticore Search

Version 28.4.4

Released: July 10th 2026

This release focuses on KNN performance, conversational search improvements, easier installation, better faceting and embedding controls, and fixes across authentication security, replication, SQL compatibility, distributed queries, and columnar/KNN internals.

New Features and Improvements

• 🆕 v28.4.0 PR #4707 Improved KNN search performance by batching distance calculations during the rescore pass, reducing overhead when final full-precision distances are recomputed and re-sorted.
• 🆕 v28.3.0 Issue #688 PR #4690 Updated Buddy to 4.2.0, adding custom_prompt support for CREATE CHAT MODEL, so conversational search answers can follow custom instructions such as citation rules and response style. It also fixes issue #686 - Buddy worker FD leak.
• 🆕 v28.2.0 PR #4596 Added a curl/wget pipe-to-shell installer in the quick start guide, making it possible to install, upgrade, and select Manticore versions with a single bootstrap command.
• 🆕 v28.1.0 Issue #683 PR #685 Updated Buddy to 4.1.0, adding HTTP JSON support for conversational search, so chat requests can be sent through the /search JSON API in addition to SQL CALL CHAT.
• 🆕 v27.4.0 PR #169 Updated MCL to 13.7.0, adding embeddings_threads to cap CPU threads used for auto-embedding inserts, ALTER TABLE ... REBUILD KNN, and text-to-vector KNN queries.
• 🆕 v27.3.0 PR #4624 Added zero-count facet bucket support for faceted search, including SQL ZEROES and JSON "zeroes": true, so max-mode facets can keep visible buckets even when their current filtered count is 0.
• 🆕 v27.2.0 PR #4608 Added CREATE TABLE ... profile='relevance', plus stored per-table defaults for ranker and boolean_mode, making it possible to give tables built-in full-text ranking and implicit keyword-combining behavior without repeating query options.

Breaking Changes

• ⚠️ v28.0.0 Issue #4667 PR #4668 Breaking change: plugin ABI version SPH_UDF_VERSION is bumped to 12 so token-filter plugins can correctly receive long dict=keywords_32k tokens instead of silently bypassing values over the legacy 126-byte limit. This breaks existing external UDF, ranker, and token-filter plugin binaries: they must be rebuilt before loading into this version, and token-filter implementations that copy token text into fixed-size buffers may also need code changes to handle much longer tokens. Existing table data and configuration remain compatible, and no index migration is required. Downgrade is possible if you also restore plugin binaries built for the older ABI and do not rely on the new long-token plugin behavior.

Bug Fixes

• 🪲 v28.4.4 Updated README.Debian.in and man pages.
• 🪲 v28.4.3 Issue #1250 PR #4623 Fixed incompatible multi-statement handling around COUNT(DISTINCT ...), preventing incorrect carry-over and rowset corruption in multi-query result processing.
• 🪲 v28.4.2 PR #4713 Fixed an authentication/authorization permission-check bypass in MySQL multi-statement execution, so each statement in a multi-statement request is validated correctly under auth.
• 🪲 v28.4.1 Issue #4705 PR #4712 Fixed authenticated replication-cluster restart recovery by persisting the stored replication user in cluster metadata, so nodes can restore and automatically rejoin without losing the configured cluster user.
• 🪲 v28.3.7 Issue #4697 PR #4698 Fixed interrupted columnar/KNN merge cleanup so temporary component files under the tracked temp prefix are removed correctly, preventing orphaned .tmp.spc.* files from breaking later table rename, attach, or drop operations.
• 🪲 v28.3.6 PR #4695 Improved DBeaver compatibility by accepting simple single-table aliases in SELECT queries, so default DBeaver table-browsing queries no longer fail.
• 🪲 v28.3.5 Issue #4682 PR #4683 Fixed the built-in Ukrainian lemmatizer so apostrophe words are normalized correctly, allowing forms like здоров'ям to match здоров'я under lemmatize_uk_all.
• 🪲 v28.3.4 Issue #4691 PR #4696 Fixed authenticated MySQL protocol startup compatibility so Connector/J and PyMySQL clients can complete native-password login flows and harmless session SET statements no longer fail under auth.
• 🪲 v28.3.3 Issue #4641 PR #4649 Fixed AOT blended-keyword handling to honor the configured blend_mode, restoring separator-stripped variants consistently for indexing and keyword extraction.
• 🪲 v28.3.2 PR #186 Updated MCL to include an explicit DOCS_PER_CHUNK build fix, preventing bundled KNN library build/open failures in the new submodule-based columnar workflow.
• 🪲 v28.3.1 Issue #4432 PR #4689 Fixed replication-cluster startup when a node was left alone in the cluster, so a remaining self-only node can recover cleanly instead of failing to start.
• 🪲 v28.1.5 PR #4687 Fixed distributed stored-field fetching to validate remote GETFIELD replies before using them, preventing malformed agent responses from being trusted.
• 🪲 v28.1.4 PR #4676 Fixed the migrated MCL submodule build and packaging workflow, restoring correct CI, package, and dependency-resolution handling for the bundled columnar library after the daemon-side submodule transition.
• 🪲 v28.1.3 PR #4663 Fixed several memory leaks and cleanup bugs across optional library loading, SSL key handling, RT dictionary cleanup, timer-thread shutdown, and facet-zeroes aggregation paths.
• 🪲 v28.1.2 Issue #4672 PR #4679 Fixed a crash when percentiles aggregations were used together with terms aggregations in the same /search request on multi-chunk RT tables.
• 🪲 v28.1.1 PR #4669 Fixed RT auto-optimization so OPTIMIZE TABLE no longer compacts a table below 2 disk chunks unless that lower cutoff is explicitly requested.
• 🪲 v28.0.2 Issue #4632 PR #4673 Fixed distributed queries with remote stored fields to fail on remote GETFIELD fetch errors instead of returning successful rows with silently empty stored-field values.
• 🪲 v28.0.1 Issue #4615 PR #4631 Fixed long SQL parse errors to preserve UTF-8 character boundaries, so invalid queries containing Cyrillic and other multibyte text no longer produce truncated or corrupted error messages.