OrgKernel

Why we included this project

When agents leave the demo and start doing real work, the question that keeps compliance teams up at night stops being whether the model can do the task and becomes who authorized it, what it was allowed to touch, and what it actually did. OrgKernel is a standalone Python library that answers exactly those questions for any agent stack that can speak its API, slotting alongside LangGraph, CrewAI, or AutoGen rather than replacing them. It gives every agent a cryptographic Ed25519 identity signed by your organization, mints short-lived execution tokens scoped to a single mission with tool allowlists and parameter bounds, and writes each action into a SHA-256 hash-chained audit log that resists silent edits or deletion. Because the scope checks and audit writes run in the execution path before each tool call, the governance is enforced, not just recommended. The SSO/SCIM integration ties agent permissions to your existing identity provider, so when a person leaves, their agents lose access with them. Teams with compliance or security obligations around agent behavior will find the replayable, externally verifiable audit trail the most useful piece.