HELM AI Kernel
A fail-closed execution firewall that sits between AI coding agents and the system, allowing, denying or escalating each action and writing a signed, offline-verifiable receipt of the decision.
HELM AI Kernel sits between an AI agent such as Claude Code or Codex, its MCP tools, shell commands and other actions, deciding to allow, deny or escalate each one before it runs, and writing a signed receipt of the decision that can be verified later. Verifying a receipt checks that it was not tampered with; trusting who signed it is kept as a separate, explicit step, checked against a workstation's own key by default or a manually pinned key for a receipt copied from elsewhere.
It is fail-closed by default: an unknown tool is escalated or denied before it dispatches, and an operator must type an explicit approval or denial for actions that require a decision. It requires no cloud account, no model API key, and no Docker, and it explicitly does not replace operating-system isolation, secret scanning or network firewalls, governing only the effects that pass through one of its own adapters, hooks or proxies.