typesense/typesense

What it solves

Typesense is a fast, open-source search engine that makes it easy to add a polished, typo-tolerant search experience to an app or website. It aims to be a simpler alternative to Elasticsearch and a cheaper, self-hostable alternative to the hosted Algolia service, so developers can get a great search feature working quickly without a lot of setup or tuning.

How it works

Typesense is a single lightweight binary (written in C++) that you run on your own server or in Docker. You define a "collection" (like a database table), add your data as JSON documents, and then query it over a simple HTTP API. It handles the heavy lifting of indexing and searching, including:

• Typo tolerance out of the box, so a search for "stork" still finds "Stark Industries".
• Filtering, sorting, faceting, and grouping results at query time.
• Vector search for similarity and semantic search using embeddings from ML models.
• Built-in semantic/hybrid search that can generate embeddings internally (e.g., S-BERT, E-5) or via external APIs like OpenAI.
• Conversational search (RAG) that returns a full-sentence answer based on your indexed data.
• Natural language search that turns free-form phrases into structured queries.
• Image search (via CLIP) and voice search (via Whisper transcription).
• Geo search, federated search across collections, JOINs, synonyms, and scoped API keys for multi-tenant apps.
• Raft-based clustering for high availability, with seamless version upgrades.

Who it's for

Developers and product teams who want to add fast, typo-tolerant search to their web or mobile apps—especially those who find Elasticsearch too complex to operate or Algolia too costly at scale. It's also useful for teams wanting semantic search, RAG-style Q&A, or image/voice search without building those systems from scratch.

Highlights

• Blazing fast: built in C++, with sub-50ms search latency; benchmarks show ~11ms average on 2.2M recipes and ~28ms on 28M books.
• Simple to operate: single binary, no runtime dependencies, easy Docker setup, and simple API.
• Flexible query-time settings: sorting, filtering, faceting, and grouping can be configured per query rather than fixed at index creation.
• Built-in AI features: semantic/hybrid search, conversational search (RAG), natural language search, image search, and voice search.
• Scales well: handles high concurrency (e.g., 250 concurrent queries/sec on a 3-node cluster) and supports clustering for HA.
• Open source with a managed cloud option (Typesense Cloud) and official clients for JavaScript, Python, PHP, Ruby, and more.