← All organizations

AI search infrastructure and vector database

Chroma

Chroma builds open-source search infrastructure that helps developers supply AI applications with relevant documents and data. Its database supports vector similarity, lexical, full-text, regex, and metadata search. Developers can run it on their own infrastructure or use Chroma Cloud, its managed serverless service. Chroma Sync automates ingestion, including chunking, embedding, and indexing GitHub repositories and web pages. Customers use Cloud for documentation search, code search, and code review agents.

Founded in 2022 by Jeff Huber, its CEO, and Anton Troynikov, Chroma combines database engineering with research into retrieval and context management. Its search architecture uses object storage with automatic data tiering and caching. The Context Rot study tested 18 language models while holding task complexity constant, finding inconsistent performance as input length increased. Its Context-1 search model applies context management during iterative retrieval: it breaks questions into subqueries, discards irrelevant results, and returns ranked supporting documents to a separate answering model.

In 2026, the company reported use in over 90,000 open-source GitHub codebases, a measure of its developer ecosystem. Chroma raised an $18 million seed round led by Quiet Capital in 2023.

www.trychroma.com

2 talks

Newest first

2 speakers at AIE

Affiliations reflect their AIE appearances, not necessarily current employment.

Start here

  1. Retrieval Augmented Generation in the Wild

    Start here for feedback-aware agent memory and Voyager’s retrievable Minecraft skills as examples of retrieval supporting ongoing agent activity.

    Anton TroynikovAI Engineer Summit 2023

  2. How to look at your data; what to look for, how to measure

    Follow Huber’s retrieval evaluation examples through a Weights & Biases chatbot and Weave traces; Jason Liu’s separate portion shows how segmenting conversation and usage data makes evaluation gaps actionable.

    Jeff Huber · Jason LiuAI Engineer World's Fair 2025

Messages from the stage

Evaluate embeddings against application queries

Huber uses generated and real queries, application-specific embedding evaluations, and recall@10 to reveal performance differences that broad MTEB rankings can obscure.

Control what enters context

Troynikov discusses context distractors alongside chunking, hierarchical retrieval, re-ranking, and irrelevant-result detection, treating production retrieval as more than a vector search operation.

Affiliations reflect each recorded session, not necessarily current employment.

Company sources · checked 2026-08-27