Chroma

Chroma is an open-source search infrastructure project built specifically for AI applications, licensed under Apache 2.0. The core database is written primarily in Rust and available on GitHub, while Chroma Cloud provides a hosted, serverless version of the same engine. The project reports over 15 million monthly downloads and more than 27,000 GitHub stars, according to the Chroma website.

What It Is

Chroma is a multi-modal search database designed to power retrieval in AI and agent applications. It supports vector (semantic similarity), sparse vector (BM25/SPLADE), full-text (trigram and regex), and metadata (filtering and faceted) search within a single system. Developers can run it locally in-memory for prototyping, self-host it on their own infrastructure, or use Chroma Cloud for a fully managed, serverless deployment. The core API is intentionally minimal — the GitHub README describes it as "only 4 functions" — making it fast to integrate into Python or JavaScript/TypeScript projects.

Architecture and Storage Model

Chroma's cloud infrastructure is built on object storage (S3/GCS) rather than traditional memory-heavy architectures. The system uses automatic query-aware data tiering and caching across hot (memory), warm (SSD), and cold (object storage) layers. The homepage notes that vectors are large — roughly 15 GB of vectors per 1 GB of text — and that object storage at ~$0.02/GB/month is far cheaper than memory at ~$5/GB/month. The engineering blog describes a custom write-ahead log (wal3) built on object storage and a push-based, morsel-driven query execution engine written in Rust. Published latency benchmarks at 100k vectors (384 dimensions) show p50 warm query latency of 20ms and p99 of 57ms.

Search Capabilities

• Vector search: Semantic similarity using dense embeddings
• Sparse vector search: Lexical search via BM25 and SPLADE (added October 2025)
• Full-text search: Trigram and regex operators
• Metadata search: Filtering and faceted search
• Collection forking: Copy-on-write dataset versioning for A/B testing and rollouts
• GroupBy: Aggregate search results by metadata keys

Open-Source Deployment Model

The core Chroma database is Apache 2.0 licensed and can be run locally or self-hosted on any infrastructure. The GitHub repository (chroma-core/chroma) is the primary distribution point, with packages available via pip install chromadb and npm install chromadb. The project publishes new tagged versions of the PyPI and npm packages on a weekly cadence (Mondays), with hotfixes released as needed. The website states Chroma is used in over 90,000 other open-source codebases on GitHub, per vendor-published claims.

Update: Chroma Cloud GA and Recent Releases

Chroma Cloud became generally available in August 2025. Since then, the changelog shows a rapid release cadence:

• Chroma Cloud Sync (March 2026): Serverless data ingestion from S3, GitHub, and the web
• Metadata Arrays (February 2026): Store arrays of strings, numbers, and booleans in metadata
• Private Networking (January 2026): AWS PrivateLink support
• Customer-Managed Encryption Keys (December 2025): BYOK encryption
• Sparse Vector Search (October 2025): First-class BM25 and SPLADE support
• JavaScript Client V3 (June 2025): Complete rewrite with reduced bundle size

The latest GitHub release is version 1.5.9, published May 5, 2026. The project's direction signals continued investment in both the open-source database and the managed cloud product, with research published on context engineering, chunking strategies, and embedding adapters.

Enterprise Deployment Options

Chroma Enterprise offers deployment flexibility beyond the standard cloud: single-tenant nodes and clusters, BYOC (bring your own cloud) within a customer VPC, on-premises, multi-region, and multi-cloud configurations. Compliance certifications listed on the enterprise page include SOC II Type 2, HIPAA, FedRAMP, SSO/SAML/SCIM, PrivateLink, and CMEK. The enterprise page lists logos for organizations including Capital One, UnitedHealthcare, Weights & Biases, Cisco, Intel, Sony, and others, though these are logo displays and not confirmed case studies unless linked.