DEV Community

Tchesse Emmanuel
Tchesse Emmanuel

Posted on

Your ai angent can exécute . But should it ?

#hf26challenge

AI agents can do much more than generate text. They can call APIs, modify code, send emails, process payments, delete files, and trigger entire workflows.

The problem? A workflow that works correctly under normal conditions can still perform an action that violates your rules.

That’s why I’m building InGravity: an open-source governance layer for AI agent actions.

Practical use cases

  1. Prevent unauthorized payments
    An AI agent prepares a payment through an API. InGravity checks the action against configured rules and can approve it, block it, or require human approval.

  2. Control AI-generated code deployments
    An agent modifies an application and triggers a deployment. InGravity can help enforce policies that determine whether the deployment action is allowed.

  3. Protect sensitive files
    An agent attempts to delete files or access protected resources. Configured policies can restrict which actions it is allowed to perform.

  4. Govern automated email workflows
    An AI agent drafts and sends emails. InGravity can enforce rules requiring human approval for sensitive messages or blocking prohibited actions.

  5. Add guardrails to n8n workflows
    Instead of relying entirely on an agent’s own judgment, developers can introduce an explicit policy check before sensitive workflow actions.

  6. Control API operations
    An agent calls an external API to modify data, change settings, or trigger an operation. InGravity can evaluate the requested action before the integration executes it.

How it works

The principle is simple:

AI agent → InGravity policy check → Decision → Controlled execution

  • approved — The action is permitted.
  • blocked — The action must not proceed.
  • pending_approval — Human authorization is required.

The integration must enforce the decision for the protection to work.

Why I’m building this

AI agents need more than instructions telling them to behave safely. Developers need explicit, testable policies governing sensitive operations.

InGravity aims to make that control accessible to independent developers and small teams, without requiring complex enterprise infrastructure.

The project is open source, and I’m looking for developers working with AI agents, n8n, APIs, and automated workflows to test it and share feedback.

GitHub: https://github.com/Tchesseludovic17-cloud/ingravity

What action in your AI workflow would you never want an agent to execute without a policy check?

Top comments (0)