Security and approvals

Intelligent software should operate within the same security model as the rest of the organization. A capability is safe not because a prompt asks it to be careful, but because the runtime enforces what the request may read, change, and spend.

Identity and permissions

Every request has an actor: a person, application, workflow, or integration. Permissions determine which workspace, entity, artifact, and tool that actor can access. The same policy should apply whether the request arrives through an API, a CLI job, an agent, or an MCP client.

Policy and approval are different

A policy makes an automatic decision: this action is permitted, denied, or requires more conditions. An approval asks a person to make a contextual decision before a consequential action continues.

Use approvals for actions that change customer commitments, money, access, production systems, or difficult-to-reverse records. The approval request should carry the proposed action, the evidence that led to it, and the options available to the reviewer.

Audit and evidence

Record access changes, policy decisions, approvals, tool calls, and external actions with the identity, time, target, outcome, and correlation identifiers needed for investigation. Audit is not a separate reporting exercise: it is part of the evidence behind a run.

You now have the system model. Continue to Get started to build your first Frontal project.