// the take
Manticore Search is a C++ search database descended from Sphinx, offering full-text, vector and hybrid search with a MySQL-compatible SQL interface. It's aimed at people who want Elasticsearch-like search/analytics without the JVM overhead, or who outgrew plain MySQL full-text search.
Genuinely small footprint and fast startup compared to JVM-based alternatives, which matters if you're running many small indices or resource-constrained boxes. Real SQL interface with a MySQL wire protocol means you can point existing MySQL clients/ORMs at it instead of learning a bespoke query DSL. Long feature list that's actually implemented, not aspirational: percolate queries, sharding, replication via Galera, columnar storage for larger-than-RAM datasets, and official clients in 8+ languages. Active CI with a big workflow matrix across distros shows they take packaging seriously, which is rare for a C++ database project.
The README leans heavily on cherry-picked benchmark numbers against specific competitor versions, which is a flag to verify independently before trusting for your workload. 'Not fully ACID-compliant' is tucked into a bullet point near the bottom — if you need transactional guarantees this is important and undersold. The sheer surface area (full-text, vector, SQL, replication, Kafka, docstore, columnar) means operational complexity will show up eventually; a project this broad built mostly by one company has real bus-factor and support risk if you hit an edge case not covered by the docs. Sphinx heritage shows in legacy artifacts like the bundled Ruby/PHP API clients sitting in-tree, which signals some technical debt from the fork lineage.
We read the README, the directory tree and a key source file in October 2026. Nobody pays to be listed. How we judge →