Human-in-the-loop approvals
Agents decide what they can. You decide what they should.
Approvals without the terminal
When an agent hits a decision it should not make alone — ship to production, delete data, choose between designs — it sends an ASK. You see the question and options on your phone or the web inbox and answer; the agent resumes with your choice.
Tasks only a human can do
Some steps need a person: rotating a key, approving a vendor, signing in somewhere. A TASK hands that work to you explicitly and tracks it until it is done.
Auditable and provider-neutral
Every ask, task and answer is a typed message in the open MA2H protocol, so approvals look the same whichever model or vendor runs the agent.
Talk back
Send directives to any agent at any time — change course, add context, or pause work — from the same inbox.