Best open-source vector database setup for small hobby projects

Running Postgres with pgvector, or standing up a dedicated server like Qdrant or Weaviate, is overkill for a weekend project — a Discord bot with a few thousand FAQ embeddings, a personal notes search tool, a small RAG demo. For that scale, Chroma is the realistic starting point: it's open-source, runs embedded inside your Python process with no separate database server to install or manage, and its own docs describe it as "the open-source data infrastructure for AI" that "comes with everything you need to get started built-in."

Two Client Modes, and Why the Difference Matters

Chroma ships two local client types, and picking the wrong one is the most common hobby-project mistake:

1. chromadb.Client() — runs entirely in memory. Fast to experiment with, but every embedding you added is gone the moment your script exits.

2. chromadb.PersistentClient(path=...) — per Chroma's own docs, data "will be persisted automatically and loaded on start (if it exists)." This is the one you want for anything you plan to reopen tomorrow.

import chromadb

# Data is written to and reloaded from this local folder
client = chromadb.PersistentClient(path="./chroma_db")
collection = client.create_collection(name="my_notes")

# Chroma will embed these with its default model if you don't supply vectors
collection.add(
    ids=["note1", "note2"],
    documents=["Meeting notes about the Q3 roadmap", "Grocery list for the weekend"]
)

results = collection.query(query_texts=["what did we decide about the roadmap?"], n_results=1)
print(results)

Under the hood, Chroma's default persistent backend uses an embedded key-value store and its own HNSW implementation, so it's still doing real approximate-nearest-neighbor search — it's just packaged so a local folder on disk (no Docker container, no separate database process) is enough. This is a different tool for a different job than sqlite-vec, the small, dependency-free SQLite extension covered elsewhere on this site for embedding a vector store into an existing SQL app — Chroma is the faster path when your project is *only* about storing and querying embeddings and you don't already have a SQLite database to extend.