Chroma
by Chroma
Open-source (Apache 2.0) vector and hybrid search database for AI, with a serverless Chroma Cloud on object storage
Chroma is an open-source, Apache 2.0 search database for AI applications that combines vector, sparse (BM25/SPLADE), full-text, regex and metadata search. It is built for developers and platform teams shipping RAG and agent memory, who want to prototype locally and then move to a serverless, usage-priced managed cloud without re-platforming.
Chroma is an open-source embedding and search database, licensed Apache 2.0 and founded in 2022 by Jeff Huber and Anton Troynikov, that has become one of the default retrieval layers for LLM applications through integrations with LangChain, LlamaIndex and agent-memory tools such as Mem0. It runs in three modes: embedded in-process (useful for notebooks and prototypes), single-node on SQLite with an in-memory HNSW index, and distributed, where a gateway, write-ahead log, compactor and query executor sit on object storage and use a SPANN/SPFresh index. The distributed engine underpins Chroma Cloud, which Chroma launched in August 2025 as a serverless, usage-metered service on AWS and GCP; the Show HN launch drew 93 points and 40 comments. Since then the vendor's changelog shows a steady run of enterprise features: Chroma Sync for serverless ingestion from S3, GitHub and the web (October 2025 to March 2026), customer-managed encryption keys (December 2025), AWS PrivateLink private networking, GroupBy and ReadLevel controls (January 2026), metadata arrays (February 2026) and an EU region (April 2026). Search covers dense vectors, sparse vectors (BM25, SPLADE), trigram full-text and regex, metadata filtering and collection forking for dataset versioning. The vendor cites Capital One, UnitedHealthcare, Mintlify and Weights & Biases as users and reports 29k+ GitHub stars. It competes with Pinecone, turbopuffer, Weaviate, Qdrant and pgvector; independent reviewers consistently rate it best-in-class for time-to-first-RAG but weaker than purpose-built alternatives for high-concurrency single-node workloads.
Platform and AI engineering teams building RAG, semantic search or agent memory who want an open-source engine they can run locally and a managed cloud with the same API.
One retrieval API from laptop prototype to serverless production, with vector, sparse, full-text and metadata search in a single query and usage-based pricing instead of provisioned nodes.
At a Glance
- Category
- Infrastructure & Cloud
- Pricing
- Usage-based, Freemium, Subscription
- Target Market
- CTOs, Enterprise Developers, Data Scientists, AI Platform Teams
- Deployment
- Open-source, Cloud-first, Self-hosted, Multi-cloud
- Founded
- 2022
- Headquarters
- San Francisco, USA
Key Features
- ✓Hybrid search API
Combines dense vector, sparse BM25/SPLADE, full-text, regex and metadata filtering in one query, avoiding a separate keyword engine.
- ✓Serverless Chroma Cloud
Managed distributed engine on object storage that scales automatically, billed on writes, storage, queries and egress rather than provisioned nodes.
- ✓Embedded and single-node modes
The same open-source database runs in-process or on one server, so prototypes need no Docker, cloud account or database API key.
- ✓Chroma Sync
Serverless ingestion that keeps collections in sync with S3 buckets, GitHub repositories and web sources without custom pipeline code.
- ✓Collection forking
Copy-on-write forks of a collection for dataset versioning, A/B testing retrieval changes, or per-branch indexes without duplicating storage.
- ✓Enterprise controls
Customer-managed encryption keys, AWS PrivateLink networking, an EU region, single-tenant and BYOC clusters for regulated deployments.
Capabilities
Use Cases
- •RAG over internal documents
Engineering teams index policies and knowledge-base content, prototype locally, then promote the same code to Chroma Cloud for production retrieval.
- •Agent long-term memory
Agent frameworks store conversation history and facts per user in inexpensive collections, retrieving relevant memories with vector plus metadata filters.
- •Code search for coding agents
Chroma Sync indexes GitHub repositories so agents can retrieve source and dependency code with regex and semantic search in the same call.
- •Multi-tenant SaaS search
Product teams give each customer an isolated database or collection, paying for storage and queries used rather than idle provisioned capacity.
Ideal For
Best For
- ✓Prototyping a RAG pipeline in minutes with an embedded, in-process database and no infrastructure
- ✓Agent memory stores and per-user or per-tenant collections that need cheap, elastic storage
- ✓Hybrid retrieval combining dense vectors with BM25/SPLADE sparse search, regex and metadata filters
- ✓Code and document search over repositories and buckets kept in sync via Chroma Sync (S3, GitHub, web)
- ✓Teams that want an Apache 2.0 engine as an exit option from a managed vector service
Not Ideal For
- ✗Self-hosted single-node deployments with heavy concurrent read/write load — independent benchmarks show pgvector sustaining higher throughput, and each index is single-threaded
- ✗Teams that need many tunable index types (IVF, DiskANN, quantisation variants) — Chroma exposes few indexing knobs compared with Milvus
- ✗Azure-first organisations wanting a native managed service — Chroma Cloud runs on AWS and GCP
- ✗Workloads needing built-in neural reranking — that still requires an external library or model
Integrations
Deployment
Market Analysis
Pros
- ✓Fastest path from zero to a working RAG prototype — no Docker, cloud account or database API key needed
- ✓Large ecosystem: ~29k GitHub stars, weekly releases, first-class LangChain and LlamaIndex integrations
- ✓Published, low usage-based cloud pricing that HN commenters called straightforward and reasonable
- ✓Enterprise controls (CMEK, PrivateLink, EU region, BYOC) added steadily through 2025-2026
Cons
- ✗Self-hosted single node is single-threaded per index and RAM-bound; high-concurrency benchmarks favour pgvector
- ✗Common footgun: the default in-memory client loses all embeddings on restart unless PersistentClient is used
- ✗Few index-tuning options and no built-in reranking compared with Milvus or Weaviate
- ✗Managed cloud arrived late (promised for end-2023, shipped August 2025) and is still catching up with longer-established managed offerings
Pricing
Open source
$0
- ✓Apache 2.0
- ✓Embedded or single-node self-hosting
- ✓Community support
Starter
$0/mo + usage
- ✓$5 free credits
- ✓10 databases
- ✓10 team members
- ✓Community Slack
Team
From $250/mo + usage
- ✓$100 credits included
- ✓100 databases
- ✓30 team members
- ✓Slack support
- ✓SOC 2
- ✓Volume discounts
Enterprise
Contact for pricing
- ✓Unlimited databases and members
- ✓Single-tenant clusters
- ✓BYOC clusters
- ✓SLAs
- ✓Dedicated support
Self-hosting is free under Apache 2.0. Chroma Cloud publishes list prices: $2.50 per GiB written, $0.33 per GiB-month stored, $0.0075 per TiB queried and $0.09 per GiB returned, on top of a $0 Starter or $250/month Team platform fee. Single-tenant, BYOC and SLAs are Enterprise-only and custom-priced.
Security & Compliance
Connect
Sources
This page was written from 8 sources, 5 on domains other than trychroma.com.
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
LiteLLM
Open-source AI gateway: 100+ LLM APIs behind one OpenAI-compatible endpoint, with cost tracking and guardrails
CIQ Fuzzball
Sovereign AI and HPC orchestration: train, fine-tune and serve models on infrastructure you control
Crusoe Cloud
Vertically integrated AI cloud — own the power, own the datacentre, rent the GPUs by the minute
Nexthop AI
Ethernet switches and network software co-designed for AI clusters, built for hyperscalers and NeoClouds