Chroma Review: Open-Source Vector Database for AI and RAG

What Is Chroma?

• Product type: a self-hosted vector/embedding database, distributed as a Python and JavaScript library plus an optional server binary — not a standalone downloadable GUI app
• Repository: github.com/chroma-core/chroma
• License: Apache-2.0, confirmed via the repository's license classifier on PyPI
• Locality: fully self-hostable — Chroma's open-source database runs entirely on your own infrastructure in every deployment mode; Chroma Cloud is a separate, optional managed/serverless offering for teams that do not want to run it themselves
• Scale: over 29,300 GitHub stars, and Chroma's own site cites more than 15 million monthly downloads of the chromadb package and usage in 90,000-plus open-source codebases — figures that reflect the package's broad download footprint rather than necessarily a count of unique production deployments
• Named users referenced on Chroma's own site include Capital One, Mintlify, UnitedHealthcare, Conduit, Propel, Cofounder, Weights & Biases, and Medwise

Chroma's Project History

• Chroma's core API was built around a deliberately small surface: create_collection, add, query, and delete, aimed at making it fast to get a working retrieval pipeline running
• The project has grown to over 29,300 GitHub stars and, per Chroma's own site, more than 15 million monthly downloads of the chromadb package as of this review
• Chroma Cloud, a separate managed/serverless offering, was introduced as an optional path for teams that want the same database without self-hosting it — Chroma's site references features like SOC 2 Type II compliance, automatic data tiering, and Bring Your Own Cloud (BYOC) for enterprise on that managed tier
• The latest chromadb release on PyPI at the time of this review is version 1.5.9, published 2026-05-05

What Can You Do With Chroma?

• Simple core API — four functions cover the core workflow: create_collection to define a collection, add to insert documents, query to search by similarity, and delete to remove entries
• Automatic embedding handling — if you feed Chroma raw text without bringing your own embeddings, it tokenizes, embeds, and indexes the documents for you using a default embedding function
• Metadata filtering — you can attach metadata to documents when adding them and filter query results by that metadata alongside semantic similarity
• Three deployment modes — in-memory/ephemeral for quick prototyping, persistent local storage so data survives process restarts, and a client-server architecture for shared, multi-process access
• Language SDKs — official Python (chromadb on PyPI) and JavaScript/TypeScript (chromadb on npm) clients
• Framework integrations — Chroma is commonly used as the vector-store backend inside LangChain and LlamaIndex RAG pipelines, paired with local embedding models (for example via Ollama) or a cloud embedding API
• Chroma Cloud (separate offering) — a managed, serverless tier that Chroma's own site describes as supporting vector, full-text, regex, and metadata search, plus sparse vector search (BM25, SPLADE), automatic data tiering, SOC 2 Type II compliance, and Bring Your Own Cloud (BYOC) for enterprise; verify current feature availability directly on trychroma.com, since these are managed-tier features and not all of them are necessarily present in the self-hosted open-source database

Usage Examples: Three Ways to Use Chroma

• Create an in-memory client and add and query documents: chromadb.Client() creates an ephemeral client, create_collection() defines a named collection, add() inserts documents with optional metadata and IDs, and query() searches by semantic similarity — Chroma tokenizes and embeds the text automatically if you do not pass your own embeddings
• Use a persistent local client so data survives restarts: Chroma's documented API includes a persistent-client mode (commonly chromadb.PersistentClient(path=...) in current releases) that writes data to local disk instead of holding it only in memory — verify the exact current constructor and arguments against docs.trychroma.com before shipping code that depends on it, since client APIs can change between releases
• Pair Chroma with LangChain as a vector store in a RAG pipeline: the typical workflow is to load and split your documents, embed them with a LangChain-compatible embedding model (local via Ollama or a cloud provider), store the vectors in a Chroma collection through LangChain's Chroma vector-store integration, and then query that store as the retriever step in a retrieval chain — check LangChain's current documentation for the exact integration package and import path, since LangChain's Chroma integration has moved between packages over time

import chromadb

client = chromadb.Client()
collection = client.create_collection(name="my_collection")

collection.add(
    documents=["This is document1"],
    metadatas=[{"source": "notion"}],
    ids=["doc1"],
)

results = collection.query(
    query_texts=["This is a query document"],
    n_results=2,
)

Install Chroma

Chroma Pricing: Is Chroma Really Free?

• Open-source database: free, Apache-2.0, no usage limits imposed by Chroma itself
• Your own costs are limited to your own infrastructure (compute and storage) plus whatever embedding provider you use, if you choose a paid cloud embedding API instead of a local model
• Chroma Cloud: a separate, optional managed/serverless offering; as of this review the site references an initial free-credit allowance for new accounts, but verify current pricing and credit terms directly on trychroma.com before budgeting around a specific figure
• No account or sign-up is required to install and use the open-source database

Chroma vs. Qdrant

Who Should Use Chroma?

Competitors and Alternatives

Common Mistakes When Evaluating Chroma

Frequently Asked Questions

Sources

Related Reading

• Local LLM Software Directory 2026 — the full catalog of local AI tools this article belongs to, including Chroma's own directory entry.
• Qdrant Review — a Rust-based, self-hostable vector database with more extensive metadata filtering, compared head-to-head against Chroma above.
• cognee Review — a knowledge-graph-based agent-memory library, for comparison against Chroma's flat vector-similarity approach.
• LangChain Review — a general-purpose LLM application framework that commonly uses Chroma as its vector-store backend.