Manticore Search

Manticore Search is a search-focused database written in C++, forked from the older Sphinx Search project and rebuilt into a general-purpose search engine with SQL, JSON, and HTTP APIs. It speaks the MySQL wire protocol directly, so existing MySQL clients and tooling can query it without a new driver, while also exposing a JSON API for application-style full-text and vector search.

Beyond classic BM25 full-text search, the codebase includes vector/embedding search support, real-time indexing, and its own authentication layer covering the API, HTTP, and MySQL protocols. The project markets itself directly against Elasticsearch, arguing for lower resource usage and simpler operations for teams that don’t need Elasticsearch’s full distributed-cluster feature set.

Distributed under GPL-3.0, Manticore Search is fully open source with a dedicated test/ suite and Docker-based test tooling, packaged for installation via its own repos, Docker, and the major Linux distributions.

What You Get

• Full-text search with BM25 ranking alongside vector/embedding search in the same engine
• A MySQL wire-protocol-compatible interface so existing MySQL clients and tools work without new drivers
• SQL, JSON, and HTTP APIs for querying and indexing, plus a dedicated authentication layer across all three
• Real-time indexing for data that needs to be searchable immediately after being written

Common Use Cases

• Replacing Elasticsearch for teams that want full-text search without running and tuning a JVM-based cluster
• Adding full-text or vector search to an existing MySQL-based application via a protocol-compatible database
• Running semantic/embedding search alongside traditional keyword search in one engine instead of two systems
• Self-hosting a search backend for a product catalog, documentation site, or log search use case

Under The Hood

Architecture Manticore Search descends from the Sphinx Search codebase and has grown into its own C++ search engine, with the core query/indexing logic in src/ (accumulator, aggregate expressions, attribute indexing, binlog for durability) and a dedicated src/auth/ subsystem implementing authentication separately for the API, HTTP, and MySQL protocols — meaning access control is enforced consistently no matter which interface a client connects through. Real-time and disk-based indexes are handled as distinct code paths, reflecting the engine’s search-database heritage rather than a general-purpose RDBMS retrofitted for search.

Tech Stack C++ with a CMake build system, MySQL wire-protocol compatibility for client interop, a JSON/HTTP API for application use, and support for vector/embedding-based search alongside classic BM25 full-text ranking. Packaged via Docker and native Linux distribution repos, with docker-compose-test.yml and a TESTING.md guide supporting the test workflow.

Code Quality A dedicated test/ directory and documented TESTING.md process, plus Docker-based test infrastructure, indicate a project with an established QA workflow rather than ad hoc testing — consistent with a search engine handling durability-sensitive operations like binlogs and real-time indexing.

What Makes It Unique Manticore combines classic full-text search, vector/embedding search, and MySQL protocol compatibility in a single C++ binary, letting teams query it with existing MySQL tooling while still getting Elasticsearch-style search capabilities — avoiding the need to run a separate JVM-based search cluster alongside a relational database.

Self-Hosting

Licensing Model GPL-3.0 licensed — the full search engine is open source and self-hostable with no license key required.

Self-Hosting Restrictions None found; the GitHub repository is the same engine described on the product website, with no separate enterprise-only binary mentioned in the README or license.

License Key Required No.