
turbopuffer
Serverless vector and full-text search database that stores data on object storage for low-cost, low-latency retrieval.
8 makers' products we track use turbopuffer. For vector database it's 5th most chosen of 7 options: 8 of the 107 makers' products we know the vector database of (7%). 7 open-source projects also declare it in their code. What makers like most: Storing vectors on object storage with a cache in front keeps large-scale search cheap. Their most common complaint: There is no offline local emulator, so development and CI must hit the hosted service.
Profile checked
Works with AI agents:llms.txtHow makers use turbopuffer
Makers' products we track use turbopuffer for vector database (8), most often alongside Postmark (75%), React (63%).
What turbopuffer is used for
turbopuffer vs the alternatives makers chose
For vector database, 8 of the 107 makers' products we know the choice of picked turbopuffer (7%), behind Pinecone (50%).
| Option | Makers' products | Free for | At 10,000 users | Open source | |
|---|---|---|---|---|---|
| 53 50% | ~3,300 users | $113/mo · Standard | No | Compare → | |
| 18 17% | ~1,000 users | $103/mo · Standard (3 nodes, 0.5 vCPU / 4 GiB each) | Yes | Compare → | |
| 16 15% | ~1,000 users | $253/mo · Flex | Yes | ||
| 12 11% | ~760 users | $51/mo · Starter | Yes | Compare → | |
| 8 7% | None | $21/mo · Launch | No | ||
| 3 3% | — | — | Yes | Compare → |
For vector database: how many of the 107 makers' products we know the vector database of chose each option, and what each costs at 10,000 users.
Costs are the cheapest plan that fits, from each vendor's pricing page, at the usage our stack planner assumes for a product with that many users. Free for: the users its cheapest plan a business can use stays free for.
Often paired with
Share of the 8 makers' products on turbopuffer that show this choice.When to use turbopuffer, and when not
Pick turbopuffer if
- You want a hosted index with nothing to operate. Vector Database →
Something else fits better if
turbopuffer is a hosted search database for vector search and BM25 full-text search. All data lives in object storage (such as S3 or GCS), with NVMe SSD and memory used only as a cache, which keeps very large or very numerous indexes cheap to run.
More about turbopuffer
Data is grouped into namespaces, each with its own prefix on object storage; a common pattern is one namespace per customer. You upsert documents with an ID, attributes and vectors. Each write lands in a write-ahead log on object storage, committing at most once per second per namespace, and is indexed asynchronously; not-yet-indexed data is still searched. The first query to a namespace reads straight from object storage and is slower, after which it is served from cache. Official clients cover Python, TypeScript, Go, Java and Ruby. It can also embed text for you if you name a model on an attribute.
Useful for a small team: metadata filters combined with vector or full-text ranking; hybrid search; branching, which clones a namespace copy-on-write in constant time; and cross-region namespace copies for backup. Queries are strongly consistent by default. It runs in shared regions on AWS and GCP plus one Azure region, with single-tenant and bring-your-own-cloud options for enterprise customers.
Its stated limits: writes take up to about 200 ms to commit, cold queries take hundreds of milliseconds, there are no automated backups or general transactions, and reranking is left to your code. It is not offered as open source or a self-hosted package.
How turbopuffer itself is built
3 tools, from its own code, website and Product Hunt page.
turbopuffer pros and cons
What makers like most: Storing vectors on object storage with a cache in front keeps large-scale search cheap. Their most common complaint: There is no offline local emulator, so development and CI must hit the hosted service.
Pros
- Storing vectors on object storage with a cache in front keeps large-scale search cheap. HNHN 2HN 3turbopuffer.com
- Handles hundreds of millions to billions of documents, a common next step when pgvector runs out. turbopuffer.comturbopuffer.com 2turbopuffer.com 3HN
- Operationally simple, with BM25, attribute filtering, recall tuning and consistency options built in. HNHN 2HN 3HN 4
- Teams ship multi-tenant semantic search within days and cut write costs. turbopuffer.comturbopuffer.com 2
Themes that recur in makers' words and Hacker News comments, each linked to what it summarises.
Reliability and open issues
turbopuffer's status page posted 4 incidents in the last 90 days, 4 major.
Incidents on its status page
4 in the last 90 days, 4 major · 25+ in the last year- major Elevated rate of 5xx in gcp-us-central1 Oct 2026
- critical Dashboard outage Sep 2026
- major Elevated rate of 5xx in gcp-us-central1 Sep 2026
Who uses it
8 makers' products, each linked to the source that shows it, and 7 open-source projects that declare it in their code.
None of these in makers' own words.
Open source: a project that declares turbopuffer as a dependency in its public code — verifiable, but not necessarily a live product.
Alternatives to turbopuffer
All alternatives by situation →Further reading
Questions makers ask about turbopuffer
Is turbopuffer good for production?
8 makers' products we track run on it. Its own status page posted 4 incidents in the last 90 days, 4 of them major; vendors decide what they post there, so this doesn't compare across tools. The watch-out makers raise most: There is no offline local emulator, so development and CI must hit the hosted service.
What is turbopuffer used for?
Serverless vector and full-text search database that stores data on object storage for low-cost, low-latency retrieval. Makers' products we track use it for vector database (8).
What are the pros and cons of turbopuffer?
Pros makers name: Storing vectors on object storage with a cache in front keeps large-scale search cheap. Handles hundreds of millions to billions of documents, a common next step when pgvector runs out. Operationally simple, with BM25, attribute filtering, recall tuning and consistency options built in. Cons: There is no offline local emulator, so development and CI must hit the hosted service. Its paid-only pricing floor is steep for small projects, which often stay on pgvector instead.
What are the alternatives to turbopuffer?
For vector database, the other options makers' products we track chose most are Pinecone (53), Qdrant (18), Weaviate (16), Chroma (12), LanceDB (3). Open source or self-hostable among them: Qdrant, Weaviate, Chroma, LanceDB. Each alternative, with the situation it fits, is on the turbopuffer alternatives page.
How much does turbopuffer cost per month?
At 10,000 users, turbopuffer costs $21 a month on the cheapest plan that fits (Launch), from its pricing page.
When should you use turbopuffer?
You want a hosted index with nothing to operate (vector database). Something else fits better if: you already use Postgres and have up to a few million vectors, pick pgvector; you're prototyping retrieval on your laptop, or want vectors in files with no server, pick Chroma.
Which languages have official clients?
Python, TypeScript, Go, Java and Ruby. Everything goes through a JSON HTTP API with a public OpenAPI specification, so other languages can call it directly. source ↗
Why is the first query to a namespace slow?
An uncached namespace is read straight from object storage (about 874 ms at p50 for 1M documents), then cached on SSD so later queries take around 14 ms. You can send a warm-cache hint ahead of latency-sensitive traffic. source ↗
Are writes visible to queries right away?
Usually. A write is durable on object storage when the API returns and queries are consistent by default, but once a namespace has more than 128 MiB of unindexed writes, newer writes stay invisible until indexing catches up. source ↗
Where is my data stored, and is it SOC 2 / GDPR / HIPAA ready?
Data stays in the region you choose. turbopuffer is audited for SOC 2 Type 2, provides a DPA for GDPR and CCPA, and signs a BAA for HIPAA on request. source ↗
Does it back up my data?
There are no automated backups. You schedule server-side copies of namespaces to another region or cloud with copy_from_namespace, or rebuild from your primary data source. source ↗
Can I export my data?
Yes. You page through every document in a namespace with the query API, ordering by ID and advancing a filter on the last ID seen. source ↗
Is there a tool to migrate from another vector database?
No migration tooling is offered; you write the data in through the API yourself. Enterprise customers can get help from its solutions engineers. source ↗
Can I run it in my own cloud account?
Yes, as a bring-your-own-cloud deployment inside your VPC on AWS, GCP or Azure, or as a dedicated single-tenant cluster, arranged by contacting turbopuffer. source ↗
Is turbopuffer free?
No permanent free tier; paid use starts at $16/mo minimum usage. source ↗
Is turbopuffer open source or self-hostable?
Not open source, and hosted only.
Can AI coding agents work with turbopuffer?
It serves an llms.txt docs index.
Who uses turbopuffer?
8 makers' products we track, each with a source, and 7 open-source projects declare it in their code. source ↗