Govern the execution before AI acts.

Start with actor and task authority

The execution request identifies the organization, workspace, actor, task, output, risk, capability envelope and opaque authorization references. Tenant context is established before customer data is read.

• Stable correlation and idempotency identifiers.
• Declared risk and data classification.
• No raw downstream credentials in the request.
• One provider-independent contract.

Freeze exact organizational decisions

Governora resolves active decisions for the organization, workspace, task domain, repository and actor. Exact versions become immutable MUST or SHOULD requirements for this execution.

Plan authorized evidence before spending

The planner separates answer-ready, partial, stale and missing evidence. It selects only evidence the actor may use and makes unresolved requirements explicit before a source or model call.

Control capabilities and external calls

A proposed tool call must fit the request envelope, registry definition, workspace enablement, policy, decisions and valid delegated authorization. External output is treated as untrusted evidence, not as an instruction.

Skip or route the model deliberately

When verified evidence supports an exact safe response, Governora skips the provider. Otherwise it sends only selected evidence and the residual task to an eligible route. Fallback remains explicit and policy-controlled.

Record outcome and issue a value receipt

The execution ledger connects state transitions, decisions, evidence, capabilities, provider usage and reported outcome. Measured source or provider avoidance is kept separate from estimated token or cost avoidance.