Self-hosted open source vector database

Get started: Official documentation ·GitHub

2. Chroma

(Best for beginners)

Best for: Python projects, local AI assistants, and small-to-medium RAG applications.

• Apache 2.0 licensed.
• Can run directly inside a Python application.
• Persistent storage on your local disk.
• No cloud service needed for local usage.docs.trychroma.com+1

Get started: Official documentation ·GitHub

3. Milvus

Best for: Large datasets, high-throughput workloads, and scalable deployments.

• Open source under Apache 2.0.
• Offers a lightweight local option called Milvus Lite.
• Can also run as a standalone server or distributed deployment.
• The self-hosted software is free; managed cloud is optional.github.com+2

Get started: Official documentation ·GitHub

4. pgvector

Best for: People who already use PostgreSQL.

• Open-source PostgreSQL extension.
• Stores vectors alongside regular relational data.
• Supports vector similarity search through SQL.
• No separate vector database subscription or additional database server is necessary if you already operate PostgreSQL.www.aquila.network+1

Get started:GitHub and installation instructions

Quick comparison

How to get started without paying for cloud

For a simple local Python setup, Chroma is particularly easy to try:

This creates a local database in the vector_data directory. Chroma's default embedding setup can generate embeddings for text, so you can experiment without setting up a separate embedding service first.docs.trychroma.com+1

One important distinction: the database can be free and local even if you later choose a paid API for embeddings or an LLM. You can also use local embedding models to avoid that dependency.

Which one would I choose?

My default recommendation is Qdrant if you are building a long-term application, and Chroma if you want the easiest way to experiment with local AI and RAG.

To narrow it down, what are you planning to use the vector database for?

1. What are you building?

RAG / chat with my documents

Semantic search over documents or products

AI agents or long-term memory

Something else

2. Where should it run?

On my personal computer

On my own server or NAS

In Docker on a machine I control

3. Approximately how much data?

Under 100,000 vectors

100,000 to 10 million vectors

Over 10 million vectors

Not sure yet

[Find my best fit]