Competitor comparison
A prompt asks once. A linked consent record governs the next run.
Personal AI agents need lightweight approvals, but approval prompts alone are poor long-term memory. When an agent resumes work, changes browser state, sends a message, or publishes a website, it needs more than a yes-or-no answer. It needs to know what evidence supported the approval and which audit trail proves the action stayed in scope.
Standalone prompts are useful but thin.
A standalone approval prompt is optimized for speed. It asks the user whether the agent may proceed right now. That works for simple, reversible actions, but it does not reliably capture evidence, expiry, excluded actions, replay links, or log ranges.
Audit linked consent keeps the prompt as the user-facing moment, then creates a structured receipt that travels with the agent's execution. The receipt can later point to the exact audit range, output link, and rollback owner.
Prompt
Best for immediate, low-risk confirmation when state is obvious and not reused.
Linked consent
Best for consequential actions where approval must remain inspectable and enforceable.
Hybrid
Use the prompt as the interface and the receipt as the durable control layer.
Where Super fits
Super can keep approval lightweight while preserving consent context. A text message AI assistant can ask quickly and still create the linked record behind the scenes.
Where replay fits
For browser actions, a computer-use cache can attach page state to the receipt. For AI website-building agents, linked consent can capture page diff, sources, URL, and rollback path.