Your team is using AI. Who's governing it?
CodexDominion routes every AI task through risk classification, approval gates, role-based access control, and a full audit trail — before any provider executes.
What you actually get
Three governed capabilities — live, today
Not a roadmap. These capabilities are running in the demo and will be running in your pilot.
Approval Gates
Medium- and high-risk tasks freeze before execution. An admin must approve or reject before any AI provider is called. Low-risk tasks proceed automatically. Every decision is logged to the audit chain.
Risk levels: low / medium / high
Gate: keyword-based risk classifier
Audit event: APPROVAL_REQUESTED / GRANTED / REJECTED
Role-Based Control
Four workspace roles — owner, admin, operator, viewer — enforced at the API layer, not just the UI. Operators submit. Admins approve. Viewers read. Boundaries hold even if the frontend is bypassed directly.
Roles: owner → admin → operator → viewer
Enforcement: JWT guard + service layer
Cross-workspace: 403 enforced server-side
Exportable Audit Evidence
Every task generates a 7-event audit chain. Download as JSON or Markdown with a SHA-256 integrity hash. Show exactly what ran, when, who approved it, and what the provider returned — in a format you can hand to a reviewer.
Format: JSON (v1.1) + Markdown
Integrity: SHA-256 hash (not a digital signature)
Events: 7-event chain per task
The governed pipeline
How It Works
Every AI task passes through the same 8-step governed pipeline. No shortcuts. No bypasses. The audit chain captures all of it.
01
User Objective
User submits a task with an optional workspace context. JWT validates identity before anything else runs.
02
Workspace Role Check
Service checks the user's role in the target workspace. Viewer? Blocked. Operator? Proceed to governance.
03
Governance Risk Check
Keyword classifier scores the objective: low / medium / high risk. Result logged to audit chain regardless of level.
04
Approval Gate
Medium and high risk tasks freeze here. Admin must explicitly approve or reject before execution continues.
05
Provider Adapter
Routes to configured provider — OpenAI, Claude, or Stub. Falls back to Stub and logs PROVIDER_FALLBACK if key is missing.
06
Scoped Memory
Retrieves prior context from user- or workspace-scoped memory before execution. Writes new memory on success.
07
Execution Result
Provider executes the task. Result stored. EXECUTION_COMPLETED or EXECUTION_FAILED written to the audit chain.
08
Audit Export
Full 7-event chain downloadable as JSON + Markdown with SHA-256 integrity hash. One click from the Command Center.
See it running
5-minute walkthrough
Watch a full governance cycle: task submission, risk classification, approval gate, provider execution, and audit export download — all in a live working environment.
The demo shows the Command Center UI, a real approval workflow with an operator and admin account, and the JSON audit export with SHA-256 hash.
Watch on YouTubeHonest about what it is
What CodexDominion 5.0 is — and isn't
What it is
- •A working governed AI control plane with live approval gates
- •Role-based access enforced at the API layer — not just the UI
- •Downloadable audit evidence with SHA-256 integrity hash
- •Real provider adapters: OpenAI (GPT-4o-mini), Claude (Haiku), Stub
- •Pilot-ready foundation for teams that need governance now
What it isn't (yet)
- •Not a compliance certification (SOC 2, HIPAA, etc.)
- •SHA-256 hash is integrity verification — not a legal digital signature
- •Not production-hardened auth (bcrypt, MFA, SSO not yet built)
- •Not a multi-tenant SaaS — single-tenant pilot deployment
- •No WebSocket real-time updates (5-second polling)
3 pilot slots available
Ready to govern your team's AI?
A focused 4-week deployment. You walk away with a working governed AI control plane, full audit trail, and a roadmap for your team.