Why ACGS exists.

The gap between a plan and an action

ACGS began with a practical question: how do you let a capable agent do useful work without giving it unlimited authority? Experiments with agents made the gap between a convincing plan and dependable execution hard to ignore.

An agent could lose context, reach beyond a task, or report completion without the evidence needed to trust the result. More autonomy made those gaps more consequential.

Authority needs an explicit place

The project’s focus moved toward the boundaries around an action: who may propose it, which resources it may affect, when review is needed, and what evidence should remain.

Those questions connect policy to implementation. A useful rule must meet the actual execution path, and a completion claim must point to something a reviewer can inspect.

Build the boundary, then examine it

ACGS explores that idea through an execution kernel, Python policy checks, multi-agent coordination, evaluation tools, and workflow examples.

The components have different contracts and levels of maturity. The website keeps their setup, evidence, and operating conditions visible so you can choose a starting point and evaluate it in your own environment.

The principles that carry through

• Make authority and scope explicit before consequential actions.
• Keep the review path separate from the proposal.
• Record evidence that corresponds to the actual decision.
• State what a test, artifact, or deployment result establishes.