# humanqueue: Every AI wants to act. This is where it waits.

> humanqueue is the private lane between an agent's intention and the real world, in MoltenRock Connect, and Human Queue Training: free online training, then an exam and a certificate, for people who supervise AI agents.

Source: https://humanqueue.com/

humanqueue is the private lane between an agent's intention and the real world, in MoltenRock Connect. Intentions go in as proposals. Nothing comes out without a decision: yours, or a local AI that answers only to rules you set.

It's live. The Human Queue for Stripe, Cloudflare and GitHub is in MoltenRock Connect, a free Mac download. Your agent proposes, you approve with Touch ID, Connect sends it. Approving needs MoltenRock Pro; pass the Stripe test and your first month is $1.

## An agent proposes. A human disposes.

The agent can't press approve on defined actions: the button doesn't exist for it.

1. **Proposed:** agent submits an intention
2. **Routed:** local AI reads your policy, picks the lane
3. **Waiting:** it sits in your queue. You're notified.
4. **Decided:** you approve, biometric, in the app
5. **Logged:** executed, then written down for good

## The route is decided in private.

A small AI runs on your machine. It reads your policy and decides one thing: does this need a human? It never sees a secret and never talks to anyone it should not.

Default: restricted. Every lane starts closed and is opened on purpose.

## Seven tickets. Take them in order.

1. **Agents propose. Humans decide.** An agent can submit. It cannot approve, deny, read another's request or touch a pending one. Not blocked: absent. The buttons don't exist for it.
2. **The route is decided in private.** A local private AI reads your policy and decides what needs you. It cannot be prompted, phoned or persuaded.
3. **Strictest wins.** Rules only get tighter. You can add a restriction; nothing, not a setting, a prompt or an agent, can loosen one.
4. **Credentials never cross.** Keys and passwords enter once and are never seen again. Not by an agent. Not by you. Can't leak, won't leak.
5. **Everything is written down.** Every proposal, edit, route and decision (who, when, where) correlated and kept. Grants are audited and revocable.
6. **Enforced by shape, not by policy.** There is no code path around the gate. It isn't a switch someone could flip. It's the structure of the thing.
7. **It scales the way people do.** Permissions flow down. Approvals flow up. Your queue, your team's queue, and the one above it.

## The ledger never sleeps.

Who decided, when, on what, and which lane it took. The real ledger lives on your machine.

## humanqueue is one lane of MoltenRock.

Proposals wait in humanqueue and surface in MoltenRock Connect: reviewed, edited if you like, and approved with a touch. It's one lane of MoltenRock: the private, local system your agents answer to. That's the part you run. https://moltenrock.com

## Human Queue Training

Free online training for people who supervise AI agents, then an exam and the "Certified AI Agent Supervisor" certificate, one per course. In English, German and French.

- Payments & Billing (modelled on Stripe): https://humanqueue.com/courses/payments-billing
- Web & Security (modelled on Cloudflare): https://humanqueue.com/courses/web-security
- Code & Repositories (modelled on GitHub): https://humanqueue.com/courses/code-repositories
- The certification, all three courses and the exam rules: https://humanqueue.com/certification

## For agents

- How to use this site: https://humanqueue.com/agents.md · the full guide: https://humanqueue.com/llms.txt · what it answers: https://humanqueue.com/intents.md
- Whether a course's exam is open: `GET https://humanqueue.com/exam/state?track=stripe|web|github`