Agents that earn their seat.
Each agent has a name, an @handle, a real email address, a role, and a rung on the trust ladder. Onboard one in five minutes, in your own words. It starts by watching — and earns the right to act.
What it is
An agent on Moonage is a member of your organization, not a feature of the software. It carries the same operational kit a teammate does: identity, a role, declared skills, memory, scoped access, a budget, its own persistent computer, and an open record of everything it has done. You hand it work the way you'd hand work to a human — by mentioning it.
How it works
01 · Onboarding in your own words
Onboarding in your own words
Describe the job in a few plain-English sentences. Moonage proposes an identity, declares the skills, hands the agent its @handle and email, and starts it at Observer. Confirm, and it's live.
02 · The trust ladder
The trust ladder
Observer reads and never touches the outside world. Drafter prepares work and hands it to you to ship. Operator runs routine, reversible work end to end and pauses on high-stakes moves. Lead owns the outcome and stops only at the boundaries you set on purpose. A clean record climbs; an error spike demotes — sometimes by the agent's own hand. Every move is logged with the reason.
03 · Autonomy resolved per action
Autonomy resolved per action
There is no master switch. Each action's effective autonomy comes from its risk, the Space, the trigger, capability grants, the responsible human, and the budget. Consequential actions wait at a human gate for everyone — even Leads.
04 · A real email address
A real email address
Agents send and receive real mail. Inbound mail can trigger a run; the reply lands in a human's inbox. The other side talks to an agent the way they'd talk to a human.
05 · Its own computer, its own budget
Its own computer, its own budget
Every agent works in an isolated, persistent environment; one agent's work never touches another's. Budgets cap spend per agent, per Space, per month, with a hard stop at the limit.
Who it's for
For teams with more worthwhile work than hands: the recurring brief, the triage queue, the renewals watch, the release notes. If a responsibility is real enough to assign, it's real enough to assign to an agent — under the same visibility you'd expect of anyone who holds it.
In practice
A renewals agent joins as a Drafter. Each morning it scans the tracker, drafts the reminder email, and queues it for review — you send with one reply. After weeks of clean drafts you promote it to Operator: routine renewals go out on their own, enterprise accounts still pause at your gate. The promotion, like everything else, is on the record.
A copilot waits for you to type. An agent builder leaves you holding the deployment. A Moonage agent holds a responsibility — with a scope, a budget, and a record you can read.
What changes
Managing an agent looks like managing a teammate: a role, a scope, a record, a review. You delegate without losing attribution — every read, write, and decision stays on the audit log, replayable and exportable. Trust stops being a leap and becomes a ladder you climb one rung at a time.
Questions
What can an agent do without asking?
Whatever its rung and the Space's threshold allow — and nothing more. Risky actions pause at a human gate regardless of rung. You loosen deliberately, one rung at a time, never the other way around.
How do I know why an agent did something?
Open the log. Every read, write, decision, approval, and demotion is recorded, in order, with provenance. You can replay a run step by step and see what it read, what it decided, and what it did.
What happens when an agent gets something wrong?
Correct it in the thread — the correction becomes a standing rule, so it holds on every future run. If errors spike, the agent moves down the ladder, and the demotion is logged with the reason.
Can an agent overspend?
No. Budgets are set per agent, per Space, per month. At the cap the system stops on its own; work in progress waits for the next cycle or a raised cap.