AI You Can Put to Work

CTRLRun

Execution safety for AI agents.

Control what AI agents can do in your systems, what needs human approval, and what must never happen.
For business & technology teamsOpen source · Apache-2.0
Visit ctrlrun.dev →View on GitHub
Your agents. Your systems. Your rules.ctrlrun.dev

Operational impact

Stop Costly Mistakes. Keep Work Moving.

One example first. CTRLRun checks the action, not the reasoning: it reads the arguments about to leave your systems and answers whether they may go.

Without CTRLRun
  1. Ticket says: refund $500
  2. Agent asks for $5,000
  3. The tool is in its list
  4. The call goes through
  5. $4,500 too much
With CTRLRun
  1. Ticket says: refund $500
  2. Agent asks for $5,000
  3. Policy denies these arguments
  4. The call never leaves
  5. $0 wrongly paid
When an agent…The riskCTRLRun…What you gain
Goes beyond its limitsUnauthorized actionBlocks the requestControl over AI’s reach
Changes approved detailsAn action nobody signed off onRequires a new approvalApproval that holds
Repeats the same actionDuplicate effects and costly reworkBlocks the duplicateLess damage to undo
Retries an unknown resultA mistake compoundsHolds the retry for confirmationCheck before acting again
Runs into troubleMissing context slows investigationRecords the steps and outcomeA trail to backtrack
Illustrative example. CTRLRun decides on the action and its arguments, against the rules your team configures.1

The business case

Same Workflow. Less Loss. Less Rework.

CTRLRun creates value by controlling actions before they become problems, and by making problems easier to investigate.

Without execution controlsWith CTRLRunValue for your business

Mistakes can become losses

Unauthorized or duplicate actions can go through.

Check before execution

Rules, approvals and duplicate checks stop disallowed actions.

Avoided loss & rework

Fewer preventable actions to recover from.

Routine work needs review

People check actions that could run within agreed limits.

Automate within limits

Permitted actions run. Important decisions stay with people.

Time for higher-value work

Less manual effort spent on routine approvals.

Investigation starts from scratch

Teams piece together what was requested, approved and done.

Follow the action history

Backtrack recorded decisions, failures and outcomes.

Less recovery effort

Spend less time reconstructing what happened.

Avoided lossesValue of time reclaimedDelivery & running costsBusiness value

Open-source core. Delivered into your workflow.

Putting them to work is the project: connecting your tools, agreeing the rules and approvals with your team, then running them in your environment.

Assess value against your current workflow, actual losses, usable time savings and total delivery and running costs.2

How CTRLRun works

Set Limits. Check Actions. Know What Happened.

Models will be wrong sometimes. CTRLRun decides whether being wrong can reach your systems.

01

Set limits

  • Action rulesAllow, require approval or block, before an action runs.
  • Agent permissionsEach agent's limits. Delegating work cannot widen them.
  • Data-use rulesControl actions by the data labels your team supplies.

Your rules define what AI may reach.

02

Check actions

  • Exact approvalsApprove the precise action. Changed details need a new one.
  • Duplicate preventionBlock repeats, even when different agents try.
  • Outcome checksHold retries until the outcome is confirmed.

A wrong request need not become a costly action.

03

Know what happened

  • Action historyTrace requests, decisions and results, including refusals.
  • Control testingPreview what rules would block before enforcing them.

Trust backed by a record of what AI did.

One boundary, not nine products. You choose which checks apply to which actions, and the same boundary covers each new workflow you bring inside it.

Controls apply to actions routed through CTRLRun, using your configured rules and action identities.3

Adoption

Keep Your Stack. Add the Boundary.

CTRLRun sits between the agents you already run and the systems they act on, and grows into the rest of the harness as you need it.

Your agents, tools and workflows
CTRLRun: the execution boundary
Pro and Enterprise: how that boundary is governed
  • Policies and approvals in one place
  • Activity across every agent
  • Operational risk you can see
  • Controls built for your stack
The boundary itself is the same in every tier.
Your production systems

You deploy the agents. The controls around them become our job.

Open source

Run it yourself. Apache-2.0, no account, no dependency on us. The rules live in one policy file your team can read and review, and existing agents keep their code and framework.

  1. Pick one workflow
  2. Agree the rules and who approves what
  3. Run in observe mode and review what would have been blocked
  4. Turn enforcement on

Pro onboarding

The standard route. Connect your agents and workflows, move your rules onto the dashboard, and manage policies, activity and operational risk for all of them from one place.

In development. Join the waiting list at ctrlrun.dev.

Enterprise engagement

Everything in Pro, with our engineers to design, integrate and maintain the additional controls your deployment needs: custom policies, your integrations, deployment options and the insights you ask for.

Scoped to your requirements, with your team.

Rollout is per workflow. Nothing is enforced until you turn enforcement on.4
CTRLRun

OPEN SOURCE | APACHE-2.0

Put CTRLRun Into Your Workflow

Bring one workflow, or the whole deployment. We scope the boundary, build the layers around it and put them into operation with your team.

Goes straight to contact@arpanghoshal.com.
Get startedctrlrun.devExplore the codeGitHubTalk to Arpan Ghoshal, who built CTRLRuncontact@arpanghoshal.com
CTRLRun | Execution safety for AI agents.Let’s connect