Chroma gives LLMs memory with ultra-low latency vector search, automatic metadata filtering, and native integration into Python & JS ecosystems.
pip install chromadb
POWERING VECTOR MEMORY FOR HIGH-THROUGHPUT AI TEAMS
Relational and keyword-based search systems struggle with high-dimensional embedding lookups. As RAG pipelines scale, queries slow down, cloud costs skyrocket, and setup complexity stalls development.
Standard SQL database vector extensions hit indexing bottlenecks when dataset sizes cross millions of embeddings.
Setting up multi-node distributed vector clusters just to prototype a RAG application wastes critical developer hours.
Manually transforming text to vectors via separate APIs introduces sync bugs, schema drift, and payload bloat.
Simple enough to run in a Jupyter notebook with `import chromadb`. Powerful enough to scale to billions of vectors in production.
Optimized C++ core utilizing HNSW indexing algorithms delivers ultra-fast cosine, L2, and inner-product distance calculations.
Combine vector similarity queries with complex boolean metadata filters (`where={"author": "alex", "year": {"$gt": 2024}}`).
Plug-and-play embedding integrations for OpenAI, Cohere, HuggingFace, and local Ollama models with zero boilerplate code.
Start completely in-memory for testing, persist to local disk with DuckDB/SQLite, or deploy to distributed Chroma Cloud seamlessly.
SOC2 Type II certified infrastructure with full end-to-end encryption at rest and in transit, API key auth, and tenant isolation.
Built-in Prometheus metrics and OpenTelemetry instrumentation to monitor indexing throughput, memory usage, and latency distribution.
Chroma handles document parsing, embedding generation, indexing, and vector similarity calculations out of the box with an intuitive, clean SDK.
# 1. Initialize local persistent database import chromadb client = chromadb.PersistentClient(path="./chroma_db") # 2. Get or create collection with metadata collection = client.get_or_create_collection( name="ai_knowledge_base", metadata={"hnsw:space": "cosine"} ) # 3. Add documents with automatic embedding generation collection.add( documents=[ "Chroma v4 provides sub-5ms vector queries", "Vector databases store high-dimensional embeddings" ], metadatas=[{"cat": "perf"}, {"cat": "tech"}], ids=["id1", "id2"] ) # 4. Perform vector similarity query results = collection.query( query_texts=["How fast is vector search?"], n_results=1 )
Zero-friction connectors for leading model providers and RAG frameworks.
text-embedding-3-large & small
First-class vector store support
Native index storage connector
Local open-weight embeddings
Open source is free forever. Chroma Cloud provides managed elasticity.
Full features for local development, research, and self-hosted deployments.
Fully managed serverless vector store with auto-scaling and zero ops.
Dedicated single-tenant clusters in your VPC or ours for high-volume apps.
"Chroma completely removed the friction of setting up a vector database for our RAG agent. We went from prototype to production in two days."
"The metadata filtering capabilities in Chroma v4 are ridiculously fast. Filtering millions of customer context vectors takes under 4 milliseconds."
"Being able to start locally in a Python notebook and migrate seamlessly to cloud infrastructure without code changes is a game changer."
Join tens of thousands of developers using Chroma to store, search, and retrieve high-dimensional AI vectors with zero friction.