Execution Boundary

Runtime

Runtime is the governed execution boundary. An intent is evaluated against active policy before an effect can occur, and the result is returned as a bounded decision that can be inspected and verified.

Decision Before Effect

Runtime keeps execution behind authorization.

The runtime boundary is where a requested action becomes a governed decision. No implicit approval is inferred from an absent or incomplete result.

1. RECEIVE

Bind the request to its declared tenant, actor, tool, and intended resource.

2. EVALUATE

Apply the active policy and return an explicit permitted, denied, or deferred outcome.

3. RECORD

Return the decision and its evidence reference before any authorized effect proceeds.

Fail-Closed Boundary

Uncertainty is a held state.

No decision, no effect

If policy evaluation, identity binding, lineage resolution, or required trust material is unavailable, Runtime does not authorize execution.

Evidence follows the decision

A governed outcome carries the references needed to inspect what was evaluated and under which policy state.

Existing systems can enter at Runtime

MCP Gateway and Runtime provide an execution-first path for governing existing AI systems without requiring adoption of every Keon component.

Human authority remains explicit

A deferred or human-gated outcome remains held until the required authority is recorded. Runtime does not infer approval from intent alone.