We are opening up autocontext because context engineering should be easier to inspect, improve, and share.
autocontext grew out of an internal need. We were building AI systems across multiple codebases and partner workflows, and every project had the same hidden cost: assembling the right context before asking a model to do meaningful work.
The naive path is to paste files, transcripts, tickets, notes, and logs into a prompt until the model has enough. That does not scale. It is hard to reproduce, hard to debug, and hard to teach to another engineer.
What autocontext is for
autocontext helps collect, structure, and preserve the context an AI system needs for a specific task. It is designed for engineering teams that want less hand-rolled glue and more repeatable context assembly.
The goal is not to replace judgment. The goal is to make judgment visible. When a system chooses certain files, documents, examples, or constraints, the team should be able to see why and refine the selection over time.
Why open source it
We use open-source infrastructure every day. We also believe context engineering is early enough that the field benefits from shared patterns. The most useful tooling will come from teams trying it against real systems, finding sharp edges, and contributing better approaches.
Open source also keeps us honest. If a tool cannot be explained clearly enough for others to use it, it probably is not mature enough to hide behind a brand.
What comes next
We will keep using autocontext inside Grey Haven projects and publish improvements as the patterns stabilize. The best version of the tool will come from production use, not from a launch post.
