Platform
Pre-execution governance for AI agents and autonomous workflows
A governed execution core decides and executes. Six capability layers sit on top of it, each with its own acceptance evidence. Every claim on this page carries a verdict, and every verdict is reviewed on a stated date.
What changes when authority is checked first
Observability, guardrails and logging all describe a decision that has already been made. Pre-execution governance is the decision itself.
After the fact
- Describes an action that already happened
- A model or heuristic decides what looked acceptable
- Logs can be written by whatever wrote the action
- Stopping a bad action depends on noticing it in time
Before it happens
- Authority is checked before the action is transmitted
- Deterministic rules over a versioned policy, not judgement
- Evidence is signed by a key the writer cannot forge alone
- A prohibited action produces no effect at all
How a governed action is decided
The same pipeline runs for every request. Two of the three possible outcomes never execute.
Identity
The acting agent is established from the authenticated credential presented at the transport, never from a field inside the request body. An agent cannot assert who it is.
Authority
Whether this agent is permitted this action, and on whose behalf, is resolved against the authority model rather than inferred from the request.
Policy
Deterministic rules are read from a versioned policy inside the same write transaction that will commit the action, so a policy change cannot land between the decision and the effect.
Decision
Allow, hold for a human, or deny. The decision is recorded as evidence in its own right, including when the answer is no.
Human approval
Where the policy requires a person, the approval is bound to that exact payload, single-use, and made by a separately authenticated human with the required role.
Execution
Exactly once, through a capability the agent never holds. The credential that performs the effect is not reachable by the requesting agent.
Signed evidence
A hash-linked record of the request, the decision, the approver where one exists, timestamps and the execution result.
Independent verification
The record is checkable offline by a portable verifier with no database, no signing key, no network and no product installation on the path.
How to read the verdicts
Proven means demonstrated from a shipped artifact against its own acceptance gate. Candidate means built and passing its own gate, but not independently accepted. Planned means nothing has been demonstrated. No verdict on this site represents a third-party assessment.