Secure Hybrid AI Hub
A local policy broker for scoped AI work, typed artifacts and verification evidence.

Open-source policy broker · 257 tests passed in September
Independent build
My experimental open-source policy broker. Real-client and end-to-end acceptance remain separate from its test evidence.
THE ENGINEERING
An experimental Python broker and CLI for governed Codex, Claude Code, VS Code and local Ollama workflows. Deterministic controls mediate scope, policy, artifacts and privileged transitions while models propose work. The September public snapshot recorded 257 passing tests with one expected private-dossier skip and Python 3.11–3.13 CI/security checks. The October review did not rerun those gates. Phase 1 closure, authority enforcement and end-to-end acceptance remain incomplete; real-client operation, browser/API delivery and distributed capacity retain separate gates.
BEFORE → AFTER
Real problems from this build, what I changed and the verified result.
- CASE 01Before
An unrestricted model could confuse a proposed action with permission to execute it.
AfterSource verification with test data is recorded; complete authority enforcement and end-to-end acceptance remain open.
How I fixed itImplemented a local broker with typed policy/task records and separately controlled privileged transitions.
- CASE 02Before
Passing tests can be mistaken for real-client production readiness.
AfterThe September public snapshot recorded 257 passes and one expected private-dossier skip; the October review did not rerun those gates.
How I fixed itSeparated published baseline tests from Phase 1 closure and real-client acceptance gates.
Evidence reviewed: Source review October 14, 2026; verification dates and open gates stated above