Search Engines

Meilisearch59.8k★

Meilisearch is the search engine behind your app's search box. Push your data in, point your frontend at it, and results appear as you type in under 50 milliseconds, with typo tolerance, filters, facets, sorting, geosearch and hybrid keyword-plus-AI search. It's the self-hostable answer to Algolia, written in Rust by a French company. The core is MIT and runs as a single binary or Docker container with no usage limits. Two things to know. Sharding and S3 snapshot streaming are Enterprise Edition code under a Business Source License, fine to test but not to run in production without a paid agreement. And it sends anonymous telemetry by default, which you can switch off. Meilisearch Cloud has no free tier, just a 14-day trial. The usage plan is $30/mo for 100K documents and 50K searches, then $0.40 per 1,000 searches; the instance plan starts around $23/mo for the smallest box. At 100K searches a month that's about what Algolia's Grow plan costs, not a fraction of it. Solo and small teams: self-host free. Teams that don't want to run it: the instance plan. The catch: search is stateful infrastructure. You own memory sizing, reindexing, backups and upgrades. typesense/typesense is the closest open source rival.

Typesense27.1k★

Typesense is the search box behind your app: type a few letters, misspell half of them, and it still finds the right result fast. It's the open source answer to Algolia, with typo tolerance, faceted filters, geo search, synonyms, and vector and hybrid search for meaning-based results. The server is free under GPL-3.0. Self-hosting is one binary or Docker image. The sizing question is RAM, since the index lives in memory: Typesense's own benchmark put 2.2 million recipes at about 900MB. Raft clustering for high availability ships free in the open source server, with a self-hosting guide. Typesense Cloud bills by the hour for a dedicated cluster sized by RAM and CPU, not by records or searches. The smallest 0.5GB node is $21.60/month, with the first 720 hours free once per account. A 3-node HA cluster at that size is $86.40/month; 8GB HA is $432/month. Solo and small teams: self-host or a small Cloud node. Larger teams: Cloud with HA, or your own cluster. The catch: RAM is the bill either way, so huge catalogs get expensive, and Algolia's personalization has no equivalent. GPL-3.0 only bites if you distribute a modified server. Meilisearch is the closest rival.

Sonic21.4k★

Sonic is a lightweight search backend that delivers fast results using a fraction of the resources of Elasticsearch. It's a search index that sits between your database and your users: you push text in, it indexes it, and it returns results fast. Written in Rust, runs in under 20MB of RAM. MPL 2.0 license. Sonic isn't a full search engine. It's a search index and suggest engine. You store your documents elsewhere (database, files) and push text to Sonic for indexing. It gives you search results as IDs, and you look up the full documents yourself. Supports autocomplete and typo tolerance. Fully free. No paid tier, no cloud hosting. Self-host only. Setup is simple: single binary, one config file, runs as a daemon. Ops burden is minimal once running. It's designed to be a background service you don't think about. Solo projects and small apps: Sonic is perfect when you need better-than-SQL search without running a JVM. Medium projects: works well up to a few million documents. The catch: Sonic doesn't do faceted search, filtering, aggregations, or relevance tuning. It's fast text search and autocomplete, nothing more. If you need those features, you need Meilisearch or Typesense. And the protocol is custom (not HTTP REST), which means you need a client library for your language.

ZincSearch17.9k★

ZincSearch does what Elasticsearch does but without the operational nightmare. It's a search engine you embed into your stack. It ingests documents, indexes them, and lets you query with the same Elasticsearch-compatible API most developers already know. The big sell: it runs as a single binary. No JVM, no cluster setup, no 4GB of RAM just to start. Written in Go, it starts in seconds and uses a fraction of the resources. For small-to-medium datasets (millions of documents, not billions), it handles search, log aggregation, and basic analytics. It's fully open source and self-hosted. No cloud offering, no paid tier. You run it yourself. Setup is trivial: download the binary, set two environment variables, run it. The built-in UI for log exploration is surprisingly good. The catch: it's not Elasticsearch at scale. If you're dealing with terabytes of logs or need distributed clustering across dozens of nodes, you'll outgrow it. The project has also seen slower development recently. For a more actively maintained alternative, Meilisearch is fantastic for user-facing search. Typesense is another strong option with a managed cloud offering if you don't want to self-host.

manticoresearch12.1k★

Manticore Search is a database built for search: the engine behind a site's search box, a log explorer, or "find by meaning" lookups for AI apps. It started in 2017 as a continuation of the Sphinx engine and now aims straight at Elasticsearch, with full-text, vector and hybrid search, and SQL as its native language. It's GPLv3 and free, with no paid edition. Setup is light. A one-line installer or a single docker run gets you going, and an empty instance sits around 40MB of memory, a fraction of what Elasticsearch wants. Replication, sharding, backups and MySQL or Postgres sync are built in. Auto-embeddings run local models free, or call OpenAI, Voyage or Jina with your key. Solo and small teams: free, and cheap to host. Larger teams: still free; the only thing for sale is support (SLAs, custom features), quoted on request. The catch: authentication is off until you turn it on, so a default install answers anyone who can reach the port. It also isn't a drop-in Elasticsearch. Only writes are compatible and Kibana support stops at 7.x, so plan on porting queries. For a simple site search box, Meilisearch or Typesense is less work.