The best vector database for you depends on whether you want vectors beside the data you already store or in a separate, purpose-built engine, and on how much accuracy you'll trade for speed. This page lists ten vector databases in no particular order, from database extensions to managed services and self-hosted engines.
A vector database stores embeddings, the lists of numbers AI models use to represent the meaning of text, images or audio, and finds the closest matches fast. It's what sits behind AI search, chat over your documents and recommendation features.
The ten options below are listed alphabetically and in no particular order. Every description comes from the provider's own website, read this week. If the idea is new, my plain guide to vector databases explains the basics first.
Vector databases at a glance
I checked the free plan column on each provider's own website, as of Oct 2026. Yes means the page I read names a free plan, free tier, free credits or open source software. Trial only means a free trial or starter credit. Not stated means the pages I read named neither, even where there's a button inviting you to start, so check the provider's page before you sign up.
| Tool | Type | Good for | Free plan |
|---|---|---|---|
| Chroma | Open-source search infrastructure | Vector plus full-text search on object storage | Yes |
| LanceDB | Multimodal lakehouse for AI | Vectors beside multimodal training data | Not stated |
| Milvus | Open-source vector database | Growing from laptop to cluster | Yes |
| MongoDB Vector Search | Vector search inside MongoDB Atlas | Vectors stored beside app data | Not stated |
| pgvector | Postgres extension | Vector search inside Postgres | Yes |
| Pinecone | Managed vector database | Search, upsert and rerank from one index | Not stated |
| Qdrant | Vector search engine | Self-hosted or managed vector search | Yes |
| turbopuffer | Search engine on object storage | Very large vector and text indexes | Not stated |
| Upstash Vector | Serverless vector index | Small projects on a free plan | Yes |
| Weaviate | Open-source AI database | Vector search, RAG and memory together | Yes |
Ten vector databases worth a look
Chroma
Best for: Developers who want open-source vector, full-text, regex and metadata search that runs locally or on Chroma's cloud.
LanceDB
Best for: AI teams that want unified vector, full-text and hybrid search with SQL filters over multimodal datasets.
Milvus
Best for: Engineers who want an open-source vector database that starts as Milvus Lite in a notebook and grows to standalone or distributed.
MongoDB Vector Search
Best for: Companies already on MongoDB Atlas who want search and vector search in one workflow next to their operational data.
pgvector
Best for: Postgres users who want open-source vector similarity search without adding a separate database to their stack.
Pinecone
Best for: Builders who want one API key to search, upsert and rerank from the same index, including from coding agents.
Qdrant
Best for: Teams that want a vector search engine for RAG, recommendations and agents, run on their own servers or in the cloud.
turbopuffer
Best for: Products with huge datasets that want vector and full-text search built on object storage so they stay scalable.
Upstash Vector
Best for: Side projects and startups that want a serverless vector index created in a console, starting on a free plan.
Weaviate
Best for: AI builders who want vector search, RAG and agent memory in one open-source, deployment-agnostic platform.
Which database fits the job
- Multimodal datasets with SQL filters: LanceDB.
- Vectors next to MongoDB data: MongoDB Vector Search.
- Very large indexes on object storage: turbopuffer.
- Side projects on a free plan: Upstash Vector.
- Search, RAG and memory together: Weaviate.
Every fast index trades a little accuracy
ANN-Benchmarks plots recall, the fraction of true nearest neighbours found on average over all queries, against queries per second for each algorithm.
Those charts show what vendors rarely put on the homepage: speed and accuracy pull against each other. Approximate search returns results quickly by checking fewer candidates, so it'll occasionally miss the best match. For most AI search that's fine. For anything where a missed document matters, ask how recall is measured, which index settings you can tune and whether you can run exact search on smaller collections.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Postgres extension, managed service or self-hosted engine?
A database extension adds vector search to the database you already run. It's the simplest stack, and your data stays in one place.
A managed vector service runs everything for you. You'll call an API and the provider handles scaling, backups and updates.
A self-hosted engine gives you a dedicated vector database on your own servers, with full control over settings, location and upgrades.
There's no wrong answer here, only a wrong order. A sensible order is to start with what you already run, prove the feature works for users and only then move to a dedicated engine. If you're building search over a few thousand help articles, you won't need the same setup as a company indexing every product photo it's ever taken.
Questions to answer before you pick one
- How many vectors will you store in a year?
- Do you need keyword search as well as vector search?
- Do you need to filter by date, user or category?
- Where must the data be stored?
- Who will run upgrades and backups?
- How will you measure whether results are right?
Do you make a vector database? Tool makers can ask to be considered for a featured place on this list.
My plain advice on vector databases
Start with the simplest option that fits your stack, and don't add a separate vector database until you've outgrown it. For many small products, vector search inside the database you already use is plenty, and it saves a whole extra system to look after.
Spend more time on what goes into the index than on the index itself. My knowledge base prep checklist covers the clean-up that makes AI search answers better.
Frequently asked questions
What is a vector database used for?
It stores embeddings and finds the closest matches, which powers AI search, chat over documents and recommendations.
Do I need a separate vector database?
Not always. A vector extension for a database you already run is often enough for smaller projects.
What is recall in vector search?
ANN-Benchmarks describes recall as the fraction of true nearest neighbours found, on average over all queries.
Are there open-source vector databases?
Yes. Several options on this list describe themselves as open source on their own sites.
Can a vector database do keyword search too?
Some can. Several options on this list mention full-text or hybrid search alongside vectors.
Is there a free vector database?
Yes, in two ways: open-source software you run yourself, and managed services that name a free plan for small projects.
