Updated
Agent building blocks

Vector Database

Store embeddings and find the most similar ones fast, for semantic search, RAG and recommendations.

For your framework or language:Next.js
Ask your AI about this, with this page as the source:ChatGPT ↗Claude ↗Perplexity ↗

Trends

LanceDB gained the most ground in package downloads: 6× against the median npm package over a year.

Public counts each month: pull requests on GitHub, package downloads, stars, Product Hunt followers. They show direction, not market share.

ToolDownloadsStarsProduct Hunt
MongoDB14.2M/wk−46% vs npm—834 followers
Upstash4.7M/wk+92% vs npm—518 followers
Faiss2.8M/wk★ 41.2k—
LanceDB1.5M/wk6× vs npm★ 11.7k—
turbopuffer863.6k/wk3.5× vs npm—0 reviews
Qdrant810.5k/wk−13% vs npm★ 35.1k156 followers
Pinecone777.1k/wk−32% vs npm—650 followers
pgvector496.8k/wk+25% vs npm★ 23.4k—
Weaviate344.8k/wk−72% vs npm★ 16.9k60 followers
Chroma257.3k/wk−4% vs npm★ 29.5k58 followers
Vespa192.5k/wk★ 7.1k—
Milvus158.3k/wk−72% vs npm★ 46.3k—
ArangoDB53.3k/wk+21% vs npm★ 14.3k—
TigerGraph2k/wk——
SingleStore671/wk——

The real choice

Inside the database you already run, or a separate vector store. pgvector keeps vectors next to your rows, with one backup and one query language, but tuning large indexes is on you. A dedicated vector store handles scale, filtering and hybrid search for you, at the cost of another service to sync data into.

What makers chose98 live products, each with a source. Not market share.

Pick by situation

Tap the ones that are you — the tools that fit light up below.
  • You already use Postgres and have up to a few million vectorspgvector
  • You're prototyping retrieval on your laptop, or want vectors in files with no serverChroma
  • You want a hosted index with nothing to operatePineconeturbopuffer
  • You need an open-source store with heavy filtering or hybrid search, self-hosted or managedQdrantWeaviate

Compare the tools

Grouped by the side of the choice they answer, not ranked. Open a row for when to use it, the trade-off and what makers say.

ToolBest for
Inside your database1
pgvectordatabase · open source · free tierApps already on Postgres that want vector search in the same database, joined with normal tables.1+86 open source—496.8k/wk+25% vs npm

Use it whenYou have up to a few million vectors and want one system to back up.

Trade-offIndex tuning and memory sizing are your job, and very large indexes can crowd out the rest of the database.

llms.txt

Used by Showpad · in 86 open-source projects.

Embedded / local-first1
Chromadatabase · open source · free tierPrototyping RAG on your laptop with a pip or npm install and no server to run.12+87 open source$51Starter257.3k/wk−4% vs npm

Use it whenYou're still figuring out whether retrieval works for your use case.

Trade-offFor production you either run its server yourself or move to Chroma Cloud.

llms.txt

Used by Clarm, Conduit, Reef, Supaboard AI, VoltAgent and 7 more · in 87 open-source projects.

Chroma makes it super easy to manage embeddings for AI apps. We love the open-source focus and how quickly it integrates into RAG pipelines.
VoltAgent, the makerSep 2026 ↗
Managed vector store2
Pineconedatabase · free tierA fully hosted index with nothing to operate, sized by usage.53+31 open source$113Standard777.1k/wk−32% vs npm

Use it whenYou want vector search without running any infrastructure.

Trade-offClosed source and cloud-only, so leaving means re-indexing somewhere else.

llms.txt

Used by AI Context Flow, Brev.io, CustomGPT, Fleak, Human Behavior and 48 more · in 31 open-source projects.

Vector database for semantic search and RAG pipelines. 200K+ vectors for CDC documents, account context memory, and knowledge retrieval in milliseconds.
Youkti, the makerSep 2026 ↗
turbopufferdatabaseVery large or many-tenant indexes where storing everything on object storage keeps cost down.8+7 open source$21Launch863.6k/wk3.5× vs npm

Use it whenYou have many separate namespaces, such as one index per customer.

Trade-offNo free tier (paid plans have a monthly minimum), and it's cloud-only.

llms.txt

Used by Kepler, Pylon, Readwise, fal, Legora and 3 more · in 7 open-source projects.

Open-source vector store2
Qdrantdatabase · open source · free tierHeavy metadata filtering alongside vector search, self-hosted from one binary or on its managed cloud.18+94 open source$103Standard (3 nodes, 0.5 vCPU / 4 GiB each)810.5k/wk−13% vs npm

Use it whenYour queries combine similarity with many filters, such as tenant, date and category.

Trade-offOne more service to deploy and keep in sync with your main database.

llms.txt

Used by AICamp, April, Byterover, Conva.AI, Lyzr and 13 more · in 94 open-source projects.

After evaluating a bunch of Vector DBs to be our internal vector DB, we finally closed on QDrant because it was the one that scaled the best and had the best price performance ratio
Conva.AI, the makerSep 2026 ↗
Weaviatedatabase · open source · free tierHybrid search that mixes keyword and vector results, with built-in modules that can create embeddings for you.16+32 open source$253Flex344.8k/wk−72% vs npm

Use it whenUsers search with both exact terms and meaning, and you want both in one query.

Trade-offMore concepts and configuration to learn than simpler stores; some advanced features need a license key.

llms.txt

Used by Cortex, Depth, Insight7, Lamatic.ai, Quantera.ai and 11 more · in 32 open-source projects.

To store all our vector embeddings, now a staple for us to build forward. Their automatic 'load balancing' on which vectors are recently used is a game changer for system optimization
Quantera.ai, the makerSep 2026 ↗

Cost: the cheapest plan that fits vectors stored 1 million vectors, queries 1 million queries, vectors written or updated 500k writes, from list prices. Try your own numbers →

Cost as you grow

Each contender's cheapest usable plan as usage rises.

$0$20,000$100,000$500,0000.10.5151050100
turbopufferWeaviateQdrantPineconeChromax: vectors stored (1,536 dimensions, about 6 gb per million) (million vectors), other usage scaled with it · cheapest usable plan at each point, list prices · try your own numbers

Who switches to what

Public pull requests on GitHub since Oct 2024 whose title says "X to Y" — real code changes, by developers in general rather than makers only. Pick a flow to see its pull requests.

Qdrant → pgvector: 9 pull requestsChroma → pgvector: 7 pull requestsChroma → Pinecone: 6 pull requestsChroma → Qdrant: 4 pull requestspgvector → Pinecone: 3 pull requestsPinecone → Qdrant: 3 pull requestsChroma 17Qdrant 9pgvector 3Pinecone 3pgvector 16Pinecone 9Qdrant 7Moving fromMoving to

Before you choose

How to approach it

If you already use Postgres, start with pgvector; most solo apps never outgrow it. Store the source text and metadata next to each vector so you can filter and re-embed later. Move to a dedicated store only when you hit a real limit — millions of vectors, slow filtered queries, or hybrid search you can't build yourself.

Common mistakes
  • Storing only vectors and IDs, then having to re-fetch and re-embed everything when you want to filter by date, user or source.
  • Forgetting to delete or update vectors when the source record changes, so search keeps returning content users already removed.
  • Adding a separate vector service to sync before trying the vector search in the database you already run.

Other options

Real choices most makers here won't need to weigh.

Decided alongside

What the 98 makers' products here chose for their other decisions.

Further reading