HELM AI Kernel

A fail-closed execution firewall for AI agents. It sits between coding agents, MCP tools and shell commands, decides allow, deny or escalate, and writes a signed receipt you can verify later.

Role
Architect
Status
Active
Links
GitHub
Links
Docs
Watch an agent keep receipts

The agent wants to

Pick an action. The policy decides and the decision is signed.

Policy

  1. AllowReading is allowed
  2. Ask a humanSending needs a human yes
  3. Ask a humanLarge payments need a human yes
  4. DenyDeleting is never allowed
  5. DenyAnything not listed is denied

Receipts

0

Verify checks every hash, link and signature again, here in your browser.

    A deny-by-default guardian, hash-chained signed receipts and an offline verifier, under Apache 2.0. If HELM cannot be reached, nothing new runs. The kernel under HELM AI OS, the operating system for hybrid teams of AI agents, people and machines.

    Related entries