Skip to main content

THE AGENT IS PROBABILISTIC. THE ACTION IS NOT.

Execution safety
for AI agents.

Control the AI agents running in your platform. Any domain. Your rules.

AI action governance, not AI agent governance.
A model decides differently every run. The action it is about to take is fixed, and can be checked before it happens.

Read the docs GitHub

Open source · one line, any action

Already running agents in production? Protect my agent →

ONE EXAMPLE / THE MODEL HALLUCINATES AN AMOUNT

The model guesses.
CTRLRun does not.

The ticket saysrefund $500

The agent asks for$5,000

Without CTRLRun
  1. Nothing checks the amount
  2. The call goes through
  3. $4,500 too much
With CTRLRun
  1. Your rule checks the amount
  2. The call never leaves
  3. $0 wrongly paid

HOW CTRLRUN WORKS

Let it run. Ask a human. Or stop it cold.

Interactive walkthrough: follow one agent action through every check →

THE SAME BOUNDARY, RUN FOR YOU

Open source is the boundary. Pro and Enterprise are how it is run.

Every tier enforces the same rules on the same actions. What changes is who runs the control plane around them, and how far the controls are shaped to your deployment.

Running it yourself stays a first-class route: the core is Apache-2.0 and nothing here is required to use it. The Pro dashboard is in development. Start with the open-source core →