Taking ChromaDB to Production

Data Persistence

# Data persists automatically with PersistentClient
client = chromadb.PersistentClient(path="/data/chroma")

# Data survives restarts — just reconnect
collection = client.get_collection("my_docs")
print(f"Loaded {collection.count()} documents")

Client-Server Mode

Run ChromaDB as a service for multi-process access:

# Start server
chroma run --host localhost --port 8000 --path /data/chroma

# Connect from any process
client = chromadb.HttpClient(host="localhost", port=8000)

Docker Deployment

services:
  chromadb:
    image: chromadb/chroma:latest
    ports:
      - "8000:8000"
    volumes:
      - chroma_data:/chroma/chroma
    environment:
      - CHROMA_SERVER_AUTH_CREDENTIALS=your-token

When to Use ChromaDB vs. Cloud Solutions

Use ChromaDB when:

• Developing and prototyping locally
• Data must stay on-premise
• Low latency is critical (no network hop)
• Dataset is under ~1M vectors

Use cloud (Pinecone, Weaviate) when:

• Scaling beyond single-machine capacity
• Need managed infrastructure
• Require cross-region availability
• Team needs shared access to indexes

Common Pitfall: Running ChromaDB Without Authentication in Client-Server Mode

By default, the ChromaDB server starts without any authentication. If you expose port 8000 publicly — even briefly — anyone can read, modify, or delete your data. Always enable token authentication before exposing the server:

environment:
  - CHROMA_SERVER_AUTH_PROVIDER=chromadb.auth.token.TokenAuthServerProvider
  - CHROMA_SERVER_AUTH_CREDENTIALS=your-secret-token
  - CHROMA_SERVER_AUTH_TOKEN_TRANSPORT_HEADER=X_CHROMA_TOKEN

Then connect from the client using:

client = chromadb.HttpClient(
    host="localhost",
    port=8000,
    headers={"X-Chroma-Token": "your-secret-token"},
)

Best Practice: Separate Ingestion and Query Clients

In production, use separate process groups for ingestion (write-heavy) and querying (read-heavy). Point both to the same persistent data directory via ChromaDB's HTTP server. This prevents a slow bulk ingestion job from blocking user-facing search queries.

Named Product Reference: Coolify + ChromaDB

If you are already using Coolify for self-hosted deployments, the ChromaDB Docker image deploys in minutes using the same one-click service workflow. Pair it with Supabase (also deployable on Coolify) to store document metadata in Postgres while ChromaDB handles the vector index — a cost-effective alternative to Pinecone for teams on a budget.