Typesense

Search that forgives a mistyped query is the point of Typesense, an engine its authors position as an open source Algolia alternative and an easier to use Elasticsearch alternative.

Written in C++ and shipped as one binary with no runtime dependencies, it runs locally or in production from a single command, and upgrading amounts to swapping the binary and restarting. A Raft-based clustering mode sets up a distributed, highly available cluster when one node stops being enough.

Search here goes well past matching strings.

• Typo tolerance: typographical errors are handled out of the box, with no configuration.
• Ranking and refinement: tunable ranking, query-time sorting, faceting, filtering, grouping and distinct results.
• Vector and hybrid search: index your own embeddings, or generate them inside Typesense with built-in models.
• Natural language search: free-form phrases become structured filters, and answers can come back as sentences grounded in your data.
• Geo, image and voice search: search around a point or bounding box, by image contents, or from a voice recording.
• JOINs and scoped keys: collections join on reference fields, and API keys can be limited to certain records.

Install from the published binaries for Linux and macOS, from the official Docker image, or run a managed cluster. Client libraries exist for several languages, and the project's own benchmark reports average search processing of 11ms on a 2.2 million recipe dataset with four vCPUs.