Chroma is an open-source search database for AI that combines vector, full-text, regex, and metadata search for RAG applications.
Chroma is open-source search infrastructure for AI applications, combining vector, full-text, regex, and metadata search in a single database built on object storage like S3. Most teams reach for it as the retrieval layer behind a RAG (retrieval-augmented generation) system, storing document embeddings and pulling back the passages most relevant to a user’s question before an LLM generates a response. It started as a lightweight embedding database you could run on a laptop and has grown into infrastructure trusted by companies like Capital One and UnitedHealthcare, which is why it remains one of the most downloaded vector databases for developers building search and RAG into production applications.
Chroma stores your data as collections of embeddings, the numerical representations of text or other data that capture meaning, alongside the original documents and any metadata you attach.
Embedding: When you add a document, Chroma converts it into a vector embedding using a model from OpenAI, Google, Cohere, Hugging Face, or another supported provider, either automatically or through one you supply.
Indexing: Those embeddings get indexed for fast similarity search, so a query returns the documents whose meaning is closest to the question, not just the ones sharing the same keywords.
Hybrid search: Chroma also supports full-text search using BM25 and SPLADE, regex pattern matching, and metadata filtering, so you can combine semantic and keyword search in the same query.
Object storage tiering: Chroma Cloud keeps hot data in memory, warm data on SSD, and everything else on cheap object storage like S3, automatically moving data between tiers based on how often it is queried.
Forking: You can fork a collection the way you would fork a Git repository, which makes it cheap to test a new embedding model or chunking strategy without duplicating your entire dataset.
RAG-powered support assistant: Store your product documentation as embeddings and retrieve the most relevant sections for each customer question before an LLM writes the answer, keeping responses grounded in real content.
Semantic code search: Index a codebase so engineers can search for functions and modules by what they do, not just by matching variable names, speeding up onboarding on unfamiliar projects.
Personalized recommendation engine: Embed product descriptions or user activity and use similarity search to surface items related to what a customer just viewed, without hand-writing recommendation rules.
Document deduplication and clustering: Use embeddings to group near-duplicate support tickets, legal documents, or research papers automatically, instead of relying on exact-match keyword search.
A/B testing a retrieval pipeline: Fork a collection to try a new embedding model or chunk size against your existing setup side by side, then roll out whichever version performs better.
Yes. Chroma is fully open-source under the Apache 2.0 licence, so you can run it locally or self-host it on your own infrastructure with no cost and no record limits. Chroma Cloud, the managed serverless option, gives new accounts $5 in free credits and then bills based on usage, with no minimum spend.
Developers use Chroma as the retrieval layer behind RAG (retrieval-augmented generation) applications, storing document embeddings and searching them for the passages most relevant to a user's question before sending that context to an LLM. It also supports full-text, regex, and metadata search, so teams can combine keyword and semantic search in one system.
Chroma's core is fully open-source and free to self-host, unlike Pinecone, which is closed-source and managed only. Compared to Weaviate, Chroma is built directly on object storage like S3, which keeps storage costs far lower for large collections, though very high-throughput workloads may still favour a dedicated self-hosted cluster.
Pinecone is a fully managed vector database built for production-scale RAG and semantic search. No infrastructure to manage — it scales to billions of vectors and charges by usage.
Vector DBWeaviate is an open-source vector database with built-in modules for embedding, hybrid search, and generative search. Self-host or use Weaviate Cloud.
Vector DBQdrant is an open-source vector database written in Rust. Fast filtered search, aggressive memory optimisation, and easy to self-host.
Vector DB
LanceDB is an open-source embedded vector database that runs in-process with no server required, built for RAG pipelines and semantic search.
Vector DBUpdates from the AI world — what shipped, what we’re using in production, and what’s worth your attention. Two emails a month, no spam.
