A runtime and platform for safer AI-agent work.

Keystone sits between your coding agents and the systems they change.

Run many agents, and spend your attention on the decisions that need you.

Choose a path

Install Keystone from a beta package, launch Keystone Studio, and check that the machine is healthy.

Connect your coding agent so Claude Code or Codex starts every session from your project's recorded decisions.

See how Keystone works and why it is designed for a language model as its user.

Get started covers installation, your first project, and connecting an agent.

Concepts explain the ideas behind the design: records, approvals, effects, memory, and tool responses.

Guides walk through daily work with project memory, context packs, and model routing.

Reference lists every Keystone command and what it is for.

Security states what Keystone enforces today and where its current limits are.