How Factory powers code search with Chroma Cloud#

Factory just announced its $50M Series B funding. They build Droids, software development agents that use AI to complete tasks. Droids need context on existing code to do their work, and to solve that, Factory syncs code repositories to Chroma Cloud so agents can search and navigate them.

Factory relies on several of Chroma's unique features to make code search fast and cost effective:

• Code constantly changes, with new commits and branches. Factory leverages collection forking to incrementally index changes in separate collections, without having to exhaustively rewrite code for each change.
• Factory’s Droids use semantic search and regex search to find features, functions, and patterns across large codebases.
• Factory runs highly parallel Droids with many concurrent readers and writers across branches and commits workloads that can spike unpredictably. Chroma Cloud is serverless on object storage and scales horizontally to deliver elastic indexing and query throughput with zero operational overhead.

Background#

Factory builds Droids, software development agents that plug into the tools you already use. Developers give them everyday tasks, from fixing bugs and shipping features to bootstrapping new projects.

Factory didn’t set out to build another prompt driven assistant. They built agents that understand the bigger picture of what a developer is working on. Droids pull context from GitHub issues, documentation, Slack conversations, and most importantly the codebase. Knowing the “why” and “where” not just the “what” lets Droids tackle real development tasks more effectively.

Search enables Droids to consider high signal context#

Chroma gives Droids flexible ways to search a codebase:

• Full text search for exact matches
• Regex search for patterns
• Semantic search to find meaningfully similar code

Agents don’t always have access to the full file system. By indexing code in Chroma, every agent gets the same search capabilities wherever it runs.

Droids are git aware: they track the current branch and commit and use that to choose the right Chroma index. Instead of reingesting the whole codebase on every change, Factory uses Chroma’s collection forking: each new collection forks from the previous one and applies only the diff, sharing unchanged data. That cuts indexing time and avoids duplicate storage costs. And because each fork is addressable by its Git hash, Droids can always pull context from the exact version of the code a developer is working on.

Droids are among the highest performing agents#

Factory’s Droids hold the top score on Terminal-Bench, an open benchmark for end to end terminal tasks. In the latest run, a Droid resolved 58.8% of tasks, surpassing every other known agent.

Terminal Bench spans dependency management, build systems, debugging, AI training, and infrastructure scripting. It evaluates not only coding ability but also planning, environment adaptation, and robustness. Droids don’t just win on isolated snippets; they excel at realistic workflows.

Factory set out to build a software development agent that goes beyond answering prompts and understands the full context of a developer’s work. By combining Chroma’s search options, git aware indexing with forking, and a design that fits existing workflows, they deliver an assistant that feels native to how engineers build software. The result: smoother day to day coding and state of the art performance, as proven by Droid’s top score on Terminal Bench.

Agents start with context. Getting the right context requires search. Chroma makes indexing easy with collection forking and provides powerful search semantic, regex, and full‑text making it ideal for code search applications.