Manticore Search 27.1.5 brings built-in auth, sharded tables, and a conversational search feature that lets you run document Q&A directly inside the database - no separate retrieval layer required.
Authentication That Doesn't Stay Behind the Scenes
Manticore now supports users, passwords, bearer tokens, and fine-grained permissions across MySQL, HTTP/HTTPS, remote agents, and replication operations. Auth is not enabled by default, but once you flip that switch, anonymous access disappears. The upgrade notes explicitly warn you to roll it out in stages: upgrade remote agents and replication peers first, then masters. A JOIN CLUSTER replaces the joining node's local auth data with the donor's - so topology management matters. This makes access control a first-class concern rather than something you handle with a reverse proxy.
Sharded Tables for Write-Heavy Workloads
Version 27.1.5 introduces sharded tables: you create a logical table, Manticore distributes inserts across shards, and the lifecycle is managed in one place. No more external sharding logic bolted on top. For large write-heavy deployments, that's a reduction in operational complexity that I'd trust more than ad-hoc sharding schemes.
Conversational Search: KNN + LLM Inside the Engine
The standout feature for the vector crowd is conversational search, exposed via CREATE CHAT MODEL and CALL CHAT. Under the hood, Manticore runs KNN on a FLOAT_VECTOR field, builds LLM context from the from='...' source columns, and keeps conversation history by conversation_uuid. It returns both the answer and the supporting sources. If you already store embeddings, this turns Manticore into a self-contained retrieval-augmented generation (RAG) pipeline - no need to wire up LangChain or a separate vector store.
Vector Builds Get Multithreaded
HNSW build and rebuild times drop significantly thanks to multithreaded index construction shipped across versions 25.14.0 and 25.15.0. Initial builds, chunk merges, and ALTER TABLE ... REBUILD KNN all benefit. Manticore also improved KNN distance calculation with AVX-512 loading, added local ONNX embeddings support, and sped up ONNX inference. The net effect: shorter wait times for auto-embedding and for large vector table rebuilds.
Better Faceting and Aggregations
Faceting gains facet_filter_mode, which lets you build e-commerce style filters that preserve selected, available, and unavailable buckets under active filtering. On the analytics side, date_histogram() now respects time zones and offsets, and Manticore adds percentiles, percentile_ranks, and median absolute deviation (MAD). These make it easier to replace OpenSearch or Elasticsearch for operational analytics.
This release line also ships searchd --check for configuration validation without startup side effects, EXIT CLUSTER for online node removal, dict=keywords_32k for indexing long machine-generated tokens, and a Ukrainian lemmatizer. 65 changelog-listed fixes address crashes, replication, and KNN edge cases.
Manticore is turning into a search engine that can also serve as a lightweight data platform - auth, sharding, and in-database conversational search are the kind of additions that reduce the number of external services you need to keep running.
Source: Manticore Search 27.1.5: Auth, sharding, conversational and faster vector search
Domain: manticoresearch.com
Comments load interactively on the live page.