Human-in-the-loop for AI agents
Run a team of AI agents as one person — and never miss the moment one of them needs you.
Your agents work while you’re away. When one hits a call only you can make — ship to production, release a payment, delete data — it rings your phone and blocks until you answer. Then it carries on.
The problem
One person, a fleet of agents, and only two bad options.
Run agents like colleagues and you deliver a team’s work alone. But you can’t be present for every moment one of them needs a decision — so you either babysit (and lose the leverage) or miss things (and lose the trust).
“They let me deliver a team’s work with one person — but I miss so much.”
How it works
Ask. Ring. Answer. The agent keeps its place.
One queue across every agent you run. No new dashboard to watch — it comes to you, only when it should.
The agent asks
At a decision only you can make, your agent calls ask_me instead of stopping or guessing. Works with any MCP-native agent stack.
Your phone rings
An unmissable alert lands — backgrounded, locked, screen dark. It shows the question and the choices, tappable in one thumb.
You tap; it continues
Your answer flows straight back into the session and the agent resumes. Miss it? It re-alerts, and catches you up on return.
Why it holds
Anyone can buzz a phone. This is the part that’s hard.
A record that stands up
Who approved what, when, and from where — a full audit trail on every decision. The governance story regulators (and your own risk team) now ask for.
Reliable when it counts
Built for answering from a train: it survives dropped connections and a crashed editor, tells “you cancelled” from “the process died,” and holds a mis-tap so Undo actually undoes.
Genuinely unmissable
A native alarm that reaches you in the background and breaks through Focus — oversight that works when nobody’s watching the screen, which is the only time it matters.
Built for how you actually work
Made for the solo operator running many.
One queue, every agent
See which of your ten needs you right now, and which are just working. Each agent named, so you know who’s asking.
Miss nothing
Answers you gave while a session was down are collected on resume — delivered once, never lost across a restart.
MCP-native
Plugs into any agent that speaks the Model Context Protocol. Not one vendor’s framework — wherever your agents run.
Yours to hold
Every agent gets its own key, revocable in one click. The client is a keyless courier — the trust lives on your relay.
Pricing
Start free. Stay for ten.
Priced for one person doing the work of many — not per seat, not per ping.
- Up to 3 connected agents
- Unlimited questions
- Native phone alarm
- 7-day history
- Unlimited connected agents
- Full decision history & audit trail
- Quiet hours & priorities
- Escalation until you answer
- Multiple approvers & SSO
- Routing & on-call schedules
- Compliance audit export
- Per-company data isolation