Pick Pinecone if you want a managed service with plugins for coding agents and a knowledge-engine layer, and you don't want to run infrastructure. Pick Weaviate if open source, freedom to self-host, built-in embeddings and an agent that writes queries for you matter more. Neither vendor's site proves a winner on speed or scale, so decide on how you want to run it.
If you've copied a vector search tutorial in the past year, there's a fair chance it uses an interface one of these two vendors is now winding down. That's not a reason to avoid either. It's a reason to read this page before you write your first line of code, because both are retiring older ways of working at the same time.
Plain advice before the detail
Go with Pinecone if your team wants a hosted service, would rather not look after servers, and likes the idea of plugins for coding agents plus a layer that turns company data into governed knowledge. Go with Weaviate if you want an open-source database you can run yourself or in its cloud, with embeddings generated inside the database and a Query Agent that turns plain questions into queries. Speed and scale claims on both sites are the vendors' own, so don't let them settle it.
Old APIs on the way out, on both sides
Here's what catches people. The managed service labels its pod-based indexes and its original Vectors API as legacy. The open-source database says its GraphQL API is deprecated and will be switched off on its cloud. So older interfaces are being retired on both sides, and plenty of blog posts and sample notebooks still use them.
There's a second twist. The managed vendor has announced pg_pinecone, open-source search for data in Postgres, which blurs its old line as a closed hosted service. The practical upshot for you is simple: when you follow a guide, check which API generation it's written against. If it uses pods or GraphQL, look for the newer route in the current docs before you build on it.
Run it yourself, or have it run for you
This is the real fork. One is presented as a managed service, with a Bring Your Own Cloud option in its docs for teams that want the data in their own account. The other describes itself as open source and deployment-agnostic, with a separate hosted cloud if you don't want to manage it. If your data has to stay on your own machines, or you want to try things locally before anyone signs anything off, open source gives you more room.
What each one does, compared
| Question | Pinecone | Weaviate |
|---|---|---|
| How you run it | Managed service, with Bring Your Own Cloud in the docs | Open source and deployment-agnostic, plus a hosted cloud |
| Search types | Full-text, semantic, hybrid and sparse-vector search, plus reranking | Hybrid search combining vector and keyword in one call |
| Embeddings | Integrated embedding listed in the docs | Built-in vector generation from text and images, no external pipeline (vendor claim) |
| Agent features | Nexus knowledge engine; plugins for Claude Code, Cursor, Copilot, Codex and Gemini CLI; an MCP server | Query Agent writes database queries from natural language; Engram gives LLM agents persistent memory |
| Many tenants | Namespaces and multitenancy in the docs | A vendor-published customer quote praises its tenant system for tens of thousands of indexes |
| Older interfaces | Pod-based indexes and the original Vectors API labelled legacy | GraphQL API deprecated and due to be switched off on its cloud |
| Getting started | Free plan stated | Open source |
Agents, embeddings and search, theme by theme
For teams building agents
Both vendors are chasing agent builders, but from different angles. The managed side ships plugins for popular coding agents and an MCP server, so a developer can wire it into their editor quickly, and Nexus is pitched as a way to compile enterprise data into governed knowledge. The open-source side puts the agent inside the database: its Query Agent writes the queries, and Engram gives agents a memory that lasts between sessions.
For embeddings
Both list embeddings inside the product. The open-source database's home page goes further and says it generates vectors from text and images with no external pipeline. If you're tired of maintaining a separate embedding step, that's worth a look, though it's the vendor's own description.
For search quality
The managed service's docs name more search types, including sparse vectors and reranking. The other's docs show hybrid search in a single call. Neither site gives you a fair head-to-head on results, so run your own queries on a sample of your data. For more on retrieval stacks in general, our guide to tools for building RAG apps is a useful companion.
Three other places to keep your vectors
If you're still weighing up the wider field, our ranked list of the best vector databases goes further. These three are the closest rivals to this pair.
Qdrant
Best for: A dedicated vector engine with managed cloud, hybrid cloud and edge options; its latest release adds BM25 hybrid search without sparse vectors.
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.
Chroma
Best for: Lightweight open-source search under an Apache 2.0 licence, covering vector, full-text, regex and metadata search on object storage.
pgvector
Best for: Teams already on Postgres who'd like vector similarity search added to the database they've got.
Do you make a vector database or AI search tool? Tool makers can ask to be considered for a featured place on this page.
Want your tool on this list? Featured spots and reviews here are paid partnerships. Here's how to partner with me.
Frequently asked questions
Is Weaviate open source?
Yes, it describes itself as open source and deployment-agnostic, and it also offers a hosted cloud if you'd rather not run it yourself.
Can I self-host Pinecone?
It's presented as a managed service. Its docs list a Bring Your Own Cloud option, and the vendor has announced pg_pinecone, open-source search for Postgres data.
Which one has built-in embeddings?
Both list embeddings inside the product. The open-source database says it generates vectors from text and images with no external pipeline.
Are any of their APIs being retired?
Yes. One labels pod-based indexes and its original Vectors API as legacy, and the other has deprecated its GraphQL API on its cloud. Check which generation a tutorial uses before copying it.
Which is faster at scale?
Neither vendor's site proves it. Both make their own claims, so test with a sample of your own data and queries.
