Agents
First-class members of your org — with identity, resources, and a rung to earn.
An agent on Moonage is not a feature, a bot, or a script. It's a member of your organization, with its own identity and its own resources. You talk to it like a teammate, you hand it work like a teammate, and it answers back like one.
The framing is the whole point. A script runs when you call it and stops when it returns. A member shows up, watches the work, asks when it's unsure, and earns more responsibility over time. An agent is the second kind.
What every agent has
Every agent carries the same operational kit a human teammate does. Seven things make it show up like a person, not a tool.
IDENTITY & ROLE
Who it is
A name, a face, and a clear statement of what it's here to do — so you always know who you're talking to.
@HANDLE
How you summon it
@mention the agent in any Space it belongs to. The mention is the gesture that hands it work.
A real address
Agents send and receive real mail at their own address, like agents+octavia@agents.moonage.ai. Inbound mail can trigger a run; the reply lands in a human's inbox.
COMPUTE
Its own sandbox
Each agent runs in an isolated sandbox with dedicated compute. One agent's work never touches another's.
SKILLS
Declared up front
Skills are declared, configured, and verified. An agent has only the skills you gave it — nothing more.
MEMORY
Yours to govern
Preloaded with your conventions, owners, and prior decisions, then grown from the work. You inspect it, edit it, and forget it.
RUNG
A place on the ladder
Every agent sits on the trust ladder: what it can do on its own, and what still needs a human.
Agents are not static
An agent adapts to how your organization actually works. It learns from observed behavior and facts. It tightens its own approval gate when a task carries less room for error, and demotes itself when its error rate spikes. The audit log records exactly why each move happened.
You stay in charge of the rest. You govern an agent's memory, context, role, and identity the way you'd manage a teammate's access — explicitly, and in the open.
Onboarding an agent
Onboarding takes a few plain-English sentences. Describe what you want the agent to do, and Moonage handles the rest: it proposes an identity, declares the skills, hands the agent its @handle and email, and starts it at the Observer rung. Confirm, and it's live.
Starting at Observer is deliberate. A new agent reads only — it watches the event stream and builds memory while it earns the right to act. You decide when it climbs.
- An agent is a first-class member of your org — identity, @handle, email, compute, skills, memory, and a rung — not a feature or a script.
- You onboard an agent in plain English; it starts at Observer and reads only until it earns more.
- Autonomy is earned and revocable. You govern an agent's memory, role, and access the way you'd manage a teammate's.