The protocol layer that makes autonomous AI reviewable before it acts.
VEX puts a control point between what the agent suggests and what actually runs — then proves it happened that way.
Actor, scope, and decision basis.
Proposed action as structured context.
Principal bound to the evidence record.
Outcome cryptographically sealed for review.
VEX changes the contract between an agent and a privileged system.
Instead of asking whether a model looked aligned after the fact, VEX places an explicit governance layer between suggested action and real consequence.
Proposal is no longer permission
VEX treats model output as a request for action, not authority to execute. That split is the foundation of governed autonomy.
Authorization moves outside the model
Scope, policy, and runtime controls are evaluated independently, reducing the agent's ability to self-expand its trust boundary.
The result becomes a reviewable record
Each action that matters is bound to context, the reasoning behind it, and the outcome — so you can reconstruct what happened later.
What an artifact looks like
Every action governed by VEX produces a cryptographically sealed Evidence Capsule. Here is a realistic example of what gets generated.
Four steps between model output and consequence.
VEX is intentionally small at the control point. Its job is to make the execution boundary explicit, enforceable, and inspectable before the system acts.
Proposal
The agent proposes an action, target, and intended effect. Proposal is captured before any privileged execution begins.
Scope check
Runtime context, resource constraints, and execution surface are checked against the declared request.
Independent authorization
Policy controls outside the agent boundary determine whether the action is admissible in the first place.
Evidence sealing
Execution outcome and decision basis are bound into reviewable evidence — including denied, escalated, or blocked paths where relevant.
VEX is the control point, not the entire stack.
ProvnAI contributes across adjacent governance work, but VEX has a narrow and valuable role: it sits where proposed action becomes real execution and makes that boundary externally governable.
The execution boundary layer where proposed action is independently authorized and sealed into evidence.
Governance framing, admissibility models, and higher-order control assumptions.
Authority, witness, and continuation primitives that complement controlled execution.
Clean lines between what is shipped and what is in pilot.
The protocol story is strongest when it is precise. This roadmap distinguishes what is available today from what we are actively piloting with design partners.
- Controlled execution surface
- Cryptographically sealed evidence
- Portable evidence artifacts
- Local verification tooling
- Enterprise deployment patterns
- Design partner integrations
- Policy and review workflows
Built for environments where AI consequence is material.
VEX is not a general-purpose chatbot feature. It is for teams that need a protocol surface around real execution, policy, and evidentiary review.
Enterprise platform teams
Teams introducing autonomous systems into production environments where privileged actions need someone to review and sign off before they run.
Public sector and regulated environments
Organizations that need AI decisions to be designed for auditability, attribution, and review under formal oversight.
Research, standards, and assurance collaborators
Groups working on formal verification, identity, admissibility, and cryptographic proof paths for agentic systems.
Secure your
autonomous AI agents
We are accepting a limited number of design partners and pilot deployments for secure agent execution. Tell us what you are building.
Enterprise & Startup Pilots
Deploy controlled execution for your agent stack. We work directly with your security and platform teams.
Public-Sector & Consortium
Standard-setting conversations for secure AI in regulated environments. EU AI Act, DORA, SOC2 alignment.
Architecture Collaboration
Formal verification, ZK proof systems, hardware-rooted identity. We support open academic and institutional collaboration.
We read every submission and reply within two business days.