Skip to content

How Keystone's agents are governed

Agents propose, people decide. Every agent has one job, guardrails, a fallback and a log you can read.

EU AI Act: Keystone is decision support

  • We never grade you for credit.
  • People approve every consequential step.
  • Every recommendation is explained, and every agent step is logged.

How we design Keystone, not legal advice.

Run ledger

Every agent run, newest first: how long it took, which guardrails acted, and who decides next. Inputs are kept as a short fingerprint, never as your data.

Loading the run ledger…
Loading

Every agent, its job and its limits

Specialised agents. Each has typed inputs and outputs, a safe fallback, and a You decide checkpoint wherever a step matters.

Loading the agent registry…
Loading