Name: Ailog - RAG as a Service Platform
Availability: InStock
Rating: 4.8 (156 reviews)

ChromaDB Setup for RAG Applications

12 novembre 2025

9 min read

Ailog Research Team

Get started with ChromaDB: lightweight, fast vector database perfect for prototyping and production RAG systems.

Why ChromaDB?

ChromaDB is the fastest way to get started with vector search:

✅ No infrastructure required
✅ Runs in-memory or persistent
✅ Built-in embedding functions
✅ Perfect for prototyping
✅ Scales to production

Installation (November 2025)

DEVELOPERbash
pip install chromadb

# Optional: For persistent storage
pip install chromadb[server]

Quick Start

DEVELOPERpython
import chromadb

# Create client (in-memory)
client = chromadb.Client()

# Or persistent
client = chromadb.PersistentClient(path="./chroma_db")

# Create collection
collection = client.create_collection(
    name="my_documents",
    metadata={"description": "RAG knowledge base"}
)

Adding Documents

DEVELOPERpython
# Add documents with automatic embedding
collection.add(
    documents=[
        "This is about machine learning",
        "Python is a programming language",
        "Vector databases store embeddings"
    ],
    metadatas=[
        {"source": "ml_book", "page": 1},
        {"source": "python_guide", "page": 5},
        {"source": "db_manual", "page": 12}
    ],
    ids=["doc1", "doc2", "doc3"]
)

Using Custom Embeddings

DEVELOPERpython
from sentence_transformers import SentenceTransformer

model = SentenceTransformer('all-MiniLM-L6-v2')

# Generate embeddings
texts = ["doc1 text", "doc2 text"]
embeddings = model.encode(texts).tolist()

# Add with pre-computed embeddings
collection.add(
    embeddings=embeddings,
    documents=texts,
    ids=["id1", "id2"]
)

Querying

DEVELOPERpython
# Simple similarity search
results = collection.query(
    query_texts=["machine learning algorithms"],
    n_results=5
)

print(results['documents'])
print(results['distances'])
print(results['metadatas'])

Advanced Filtering

DEVELOPERpython
# Filter by metadata
results = collection.query(
    query_texts=["python"],
    n_results=10,
    where={"source": "python_guide"}  # Only from python_guide
)

# Filter by ID
results = collection.query(
    query_texts=["databases"],
    where_document={"$contains": "vector"}  # Document must contain "vector"
)

Updating Documents

DEVELOPERpython
# Update existing document
collection.update(
    ids=["doc1"],
    documents=["Updated content about ML"],
    metadatas=[{"source": "ml_book", "page": 2, "updated": True}]
)

# Delete documents
collection.delete(ids=["doc2"])

Production Setup (2025)

ChromaDB server mode for production:

DEVELOPERbash
# Start server
chroma run --host 0.0.0.0 --port 8000

DEVELOPERpython
# Connect from client
import chromadb

client = chromadb.HttpClient(
    host="localhost",
    port=8000
)

Docker Deployment

DEVELOPERyaml
# docker-compose.yml
version: '3.8'
services:
  chromadb:
    image: chromadb/chroma:latest
    ports:
      - "8000:8000"
    volumes:
      - chroma_data:/chroma/chroma
    environment:
      - ALLOW_RESET=true

volumes:
  chroma_data:

Performance Tuning

DEVELOPERpython
# Batch operations for speed
batch_size = 100
for i in range(0, len(docs), batch_size):
    batch = docs[i:i+batch_size]
    collection.add(
        documents=batch,
        ids=[f"doc_{j}" for j in range(i, i+len(batch))]
    )

Hybrid Search (ChromaDB + BM25)

DEVELOPERpython
from rank_bm25 import BM25Okapi

# BM25 for keyword search
tokenized_docs = [doc.split() for doc in documents]
bm25 = BM25Okapi(tokenized_docs)

# Query both
def hybrid_search(query, n_results=5, alpha=0.7):
    # Vector search (ChromaDB)
    vector_results = collection.query(
        query_texts=[query],
        n_results=n_results*2
    )

    # Keyword search (BM25)
    bm25_scores = bm25.get_scores(query.split())

    # Merge results
    # ... (combine scores with alpha weighting)

    return top_results

ChromaDB vs Alternatives (Nov 2025)

Feature	ChromaDB	Pinecone	Qdrant
Setup	Instant	Cloud signup	Docker
Cost	Free	$70/month+	Free/self-hosted
Scale	1M+ vectors	Billions	Billions
Best for	Prototyping	Production	Advanced features

Migration to Production DB

When outgrowing ChromaDB:

DEVELOPERpython
# Export from ChromaDB
chroma_docs = collection.get()

# Import to Pinecone/Qdrant
for doc, emb in zip(chroma_docs['documents'], chroma_docs['embeddings']):
    production_db.upsert(doc, emb)

Best Practices

Use persistent mode for important data
Batch operations for performance
Index metadata fields you filter on
Monitor collection size (ChromaDB is best < 10M vectors)
Backup regularly if using persistent mode

ChromaDB is perfect for getting started. It's what we use at Ailog for development and small deployments.