Back to catalog
Typesense
GPL-3.0Typesense is an open-source, in-memory search engine replacing Searchspring and Algolia with typo-tolerant sub-50ms search and dynamic runtime sorting.
Open-source alternative to — compare it against:
Overview
Typesense was built from scratch in C++ to hold search indices in memory while persisting JSON documents with RocksDB. Pairing an Adaptive Radix Tree with typo tolerance, the engine serves queries with sub-50-millisecond response times. Teams deploy it as a single binary or container to power instant search without JVM overhead.
Why Typesense instead of Searchspring
Searchspring operates as a hosted service for e-commerce merchandising, routing queries over the internet to managed cloud clusters. That architecture introduces wide-area latency and recurring fees scaled by SKU counts and search volume. Typesense replaces that remote loop with an in-memory search daemon running inside your private network. Colocating search with application servers eliminates network hops, maintains sub-50-millisecond response times, and keeps catalog traffic private without third-party API rate limits.
The divide centers on architectural control and resource efficiency. Searchspring pairs closed ranking algorithms with visual merchandising dashboards, but conceals infrastructure controls. Typesense gives engineers programmatic control over field weights, typo tolerance, dynamic runtime sorting, and vector embeddings in a single collection. While Searchspring and Algolia often require separate replica indices for each sort order, Typesense evaluates sorting dynamically at query time from one index. In return, teams manage their own infrastructure, provisioning sufficient RAM for active indexes and operating internal Raft nodes.
What it does well
- Dynamic sorting: evaluates price, date, or numeric fields on the fly within one collection, avoiding duplicate index replicas.
- In-memory index: stores tokens in an Adaptive Radix Tree to deliver sub-50-millisecond typo-tolerant matches and facet counts.
- Hybrid search: indexes precomputed vectors or uses embedded models to combine dense vector similarity with keyword search.
- Single binary: runs as a self-contained C++ executable with embedded RocksDB, removing JVM and database dependencies.
- Raft clustering: manages multi-node replication, sharding, and automated failover internally without external coordinators.
- Scoped search keys: produces cryptographically signed API tokens restricted by collection, action, or document filters.
- InstantSearch support: connects directly to frontends built with InstantSearch libraries via an official adapter.
