Buying snapshot
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Decision point | What Typesense states |
|---|---|
License | Typesense Server is GPL-3.0. The client libraries you build into your own code use the Apache license. |
Self-hosting | A single binary with no runtime dependencies. Packages for Linux (x86_64 and arm64) and Mac (x86_64), RPM and DEB packages, and an official Docker image. |
Typesense Cloud | A dedicated cluster with no limits on records or operations. Locations include N California, N Virginia, Sao Paulo, Frankfurt, Cape Town, Mumbai, Tokyo and Sydney. |
Clients and integrations | Official clients for JavaScript, PHP, Python and Ruby, community clients such as Go, .Net, Java and Rust, integrations such as Laravel, WordPress and WooCommerce, and an InstantSearch.js adapter for search interfaces. |
Support | Community help through Slack and GitHub issues. Paid support adds guaranteed SLAs, calls, private Slack, scaling guidance and prioritized feature requests. |
Our take
Choose Typesense when search is a feature your users see, such as search-as-you-type with filters and facets. Self-host the open-source server if you already run your own infrastructure; choose Typesense Cloud if you would rather pay by the hour for a dedicated cluster than operate one.
Evaluate Typesense
Comparisons
Where it stands out
Relevance set in each query. Fields to search, facet and group by, and ranking can be set through query parameters at search time, and sorted results come from one index. Typesense says this helps reduce memory use.
Keyword, vector and AI search in one engine. Typesense can index your own embeddings or generate them with built-in models such as S-BERT and E-5, or through OpenAI or PaLM API. It also lists conversational search with built-in RAG, natural language queries turned into filters and sorts, image search with CLIP and voice search through Whisper. Typesense Cloud offers GPUs for embedding generation in select configurations.
Features for multi-tenant apps. Scoped API keys limit access to certain records, JOINs connect collections at query time, and federated search covers several collections in one HTTP request. Geo search, synonyms and curation of fixed positions are built in.
Light to run. Raft-based clustering gives high availability, an upgrade means swapping the binary and restarting, and a fresh server uses about 30 MB of memory.
Help for teams leaving Algolia. Typesense ties free migration help to your current Algolia spend. Above $1,000 a month: a 30-minute consulting call. Above $2,000 a month: a year of Prioritized Business Support and Migration Consulting, with benchmarking help. Above $3,000 a month: a year of Prioritized Enterprise Support and Migration Consulting, with benchmarking help and private Slack Connect support.
Product screenshot
Trade-offs to weigh
RAM is the sizing decision. Memory use grows with the number and type of fields you index. Typesense reports about 900 MB of RAM for 2.2 million recipes and about 14 GB for 28 million books, and 165 MB in memory for 1 million Hacker News titles with their points, data that takes 88 MB as JSON on disk. Size memory against your own records before you choose a cluster.
GPL applies if you ship changes. Typesense says GPL allows running the server as a separate service, and that it avoided AGPL because AGPL is what turns network-accessed server software into derivative work. GPL obligations start when you modify the server and distribute the changes, so review them first if you plan to do that.
No personalization or server-side search analytics yet. Typesense's README says it does not currently have them. For analytics, it suggests instrumenting search on the client side and sending metrics to your web analytics tool.
Compliance paperwork needs a paid support plan. SOC2 reports, HIPAA BAAs, custom vendor forms, security questionnaires and custom contracts are not part of the free migration support and require buying an appropriate support plan.
Pricing and plan conditions
The open-source server is free to use; you pay for your own infrastructure. Typesense Cloud does not charge by records or search operations. You pay a fixed hourly cost plus bandwidth charges, depending on the cluster configuration you choose, and you can start with a free cluster. The options below shape that configuration.
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Configuration option | What Typesense says |
|---|---|
Memory | 0.5 GB to 1024 GB, based on the size of your data. The configuration assistant suggests a starting size from your record count and average record size, counting only indexed fields. |
vCPUs per node | More vCPUs support higher traffic. Start with the lowest option for your RAM and use at least 4 vCPUs above 20 requests per second. Burst vCPU options can save costs with spiky traffic; one listed option is 2 vCPUs with 1 hour of burst per day. |
High Availability | 3 to 7 nodes. Recommended for production. |
Search Delivery Network | Like a CDN for search: off, 3 regions or 5 regions. Recommended when your users are spread across several cities. |
High Performance Disk | Recommended for records over 1KB. |
Public signals
About these signals
Verified factual signals from public sources. They indicate observable activity or interest, not total adoption, product quality, or cost.
103 GitHub commits 90d27.1k GitHub stars0 vulnerabilities across 2 packages
See all signals from 10 sources
npm · typesense@3.2.0 · PyPI · typesense@2.1.1
Alternatives to Typesense
The reviewed substitutes for Typesense among the search engines, and what would make each one the better answer.
Direct alternatives
Reviewed substitutes: products bought for the same job, where a team picks one.
- Marqo
- Typesense and Marqo both power the search box of an online store or app with typo-tolerant, instant results and semantic matching. Typesense is an open-source search engine, self-hosted or on Typesense Cloud, whose ranking and facets the team configures, while Marqo is a hosted platform that learns ranking from the store's catalog and shopper clicks, carts and purchases.Applies to: Site and product search for an online store or app.
Other approaches
A different approach to the same problem. Each substitutes only for the workload named beside it.
- Pinecone
- Choose Pinecone if you need a zero-ops vector database with enterprise-grade security and serverless auto-scaling for production RAG or recommendation systems.Applies to: Whether embedding retrieval needs a dedicated vector store or a search engine that also does vectors.
- Milvus
- Choose Milvus if you need a battle-tested open-source vector database that scales to billions of vectors with GPU support.Applies to: Deciding how the stack is shaped, where both products can be part of the answer.
- Elasticsearch
- Both serve keyword and faceted search; Typesense trades Elasticsearch's breadth for typo-tolerant search-as-you-type and simpler operation.Applies to: Keyword and faceted search behind a product's search box.