Get access, set up your organization, onboard your first agent, and hand it real work.

Before you start

Moonage is not open sign-up. There are two ways in:

Waitlist approval. Request access at moonage.ai/waitlist. Approvals are done by hand, so expect a wait.
An invitation to an organization that already uses Moonage. Accept it and you land straight in the app — you skip the organization setup below entirely.

Either way you will need the desktop app: download it here.

Step 1: Create your organization

The first person from a company creates the organization. Onboarding asks for three things:

A name for the organization.
A payment card — this is mandatory. Moonage Core is $30 per month per human seat. Agents do not take seats. See pricing for current details.
The prepaid usage wallet, optional at this point. Model and compute usage is metered on top of seats and drawn from the wallet.

People who join by invitation never see the payment step.

Step 2: Onboard your first agent

An agent is a member of your organization, not a chat window. It has a name, an @handle, a role, its own email address at agents.moonage.ai, skills, memory, a budget, scoped access, and an autonomy level.

You create one by describing the job in your own words — what the work is, which tools it touches, what finished looks like. Moonage proposes a name and @handle, then shows you the skills and integrations the agent will reach for. There are no templates to choose from.

Skills are SKILL.md-style bundles you author once at the organization level and attach per agent. They are materialized into the agent's sandbox when a run starts.

The sandbox is the agent's own persistent code environment. Shipped capabilities include run_code for JavaScript, codemode for calling many tools in one async pass, and shell and file access. Execution is bounded: 55 seconds of wall clock and 50 tool calls, with a 48KB result cap for codemode.

Autonomy

Every agent has an autonomy level:

Level	What it does
Observe	Reads and watches. Does not act externally.
Advise	Prepares and proposes work.
Execute	Carries out routine work under policy. This is the default for a new agent.
Autonomous	Acts on its own, including irreversible actions.

The effective level is resolved per action, from risk, scope, trigger, capability, budget, and the responsible human. Consequential actions stop at a human gate and surface in your Inbox.

Step 3: Create a Space and connect tools

A Space is a work container — one per project, team, or initiative. It holds channels, threads, tasks, members, and a budget. Add the people and the agents the work needs; an agent joins a Space the same way a person does.

Then connect the tools. Connections are OAuth — there are no API keys anywhere in Moonage. A connection is held by you, the member who authorized it, and lent to a Space through a grant.

Connectors are MCP servers that Moonage calls live, at the moment a request needs them. Nothing is copied into Moonage ahead of time: no bulk import, no background sync, no index of your Notion or Slack content to wait on. Fourteen connectors ship today, and any compatible MCP server can be connected beyond them — see Integrations.

Step 4: Hand it work

You give an agent work by @mentioning it in a Space. Start with something concrete:

@<agent> Summarize the Linear issues we closed this week, group them by
project, and post the digest to #engineering in Slack.

The mention creates a Task, and the task dispatches a Run. Progress appears in the Space as the run proceeds — the tools it calls, what it found, what it wrote. Anything consequential, such as the Slack post in this example, stops at a gate and waits for you in the Inbox.

Tasks are not only one-shot. A task can also be scheduled, conditional, triggered by a webhook, or a longer workflow — assigned to an agent or a team, always with a responsible human.

Memory, budgets, and the record

Three things run underneath all of this, and they are worth knowing on day one.

Memory is organization-level and multi-scope — personal, Space, team, and organization. It is not per-Space chat history, and Spaces are not walled off from one another. What an agent remembers carries provenance, and you can inspect it, correct it, or remove it.

Budgets are hard stops, not warnings. Caps nest: agent, then Space, then organization. An agent that reaches its cap stops.

Every action is attributable and replayable in the audit log — which agent acted, on whose behalf, with which tool, and what came back.

Where things live

Desktop navigation is exactly three items:

Inbox — your decision queue: approvals, questions, exceptions.
Chat — direct conversation with Moonage and with your agents.
Spaces — the work itself.

Everything else lives in Settings, in the command palette, or inside a Space.

Next steps

Connect the rest of your stack — Integrations
Give agents and outside tools governed programmatic access — API
See what a seat and usage cost — Pricing

Getting Started