Getting started as a member

Go from an invite link to an agent working for you in your own space. A space is a private, isolated environment where your agent, your code and your tools run together. By the end of this page you will have given an agent its first task and watched what it did.

Runs and deputies are available when your administrator enables them.

Accept your invite

Open the invite link your admin sent you. The page says “You’re invited” and shows the organization name and your role.
If you already have an account, select Accept & switch to, followed by the organization name.
If you do not, select Create account. Enter your Name (optional), Email address and Password, then select Create account.

What you should see: you land on Home, and the top bar shows the organization name and your role.

Each link works once and is valid for 7 days. If the page says the invitation expired, was used, or was revoked, your admin sends you a new one from Settings, then Members.

Next time, open your Marshal address and select Sign in, or Continue with Google if your organization uses it. Forgot password? on the login page gets you back in.

Add your credentials

A space uses credentials to reach services, such as your agent account and Git. You add your own, once. Nobody can read the saved value, including admins, and the agent never sees it.

Open Connections in the left rail.
Select Connect on the card you need.
Follow the dialog. Many dialogs have one tab for each way to sign in.
Leave Owner as Just me. You can share a credential later.
Select Save, or the connect button for browser sign-in.

For Claude Code, paste an API key (it starts with sk-ant-api, sk-ant-svc or sk-ant-usr), or run claude setup-token on your own machine and paste the token (it starts with sk-ant-oat). For GitHub, sign in with GitHub, or use a personal access token or an SSH key. An SSH key gives Git access only, so the agent cannot open pull requests with it.

What you should see: the credential appears on Connections with a green status dot.

Full steps: Add a Claude credential and Connect GitHub.

Create your first space

Select the rocket button Create a new space at the bottom of the left rail. You can also open Spaces and select New Space.
Under Manifest, select Choose a manifest… and pick one. Your admin usually gives every member one.
Enter a Space name if you want one. Marshal generates a name if you leave it empty.
Select Create Space.

What you should see: the page “Starting space” with a progress card. When it finishes you land in the space and see “Ready to connect”.

If the manifest needs a credential you have not added, the page tells you which one. Add it on Connections and try again. Full steps: Create a space.

Give the agent its first task

Select Open Terminal. A terminal opens in your browser, with the agent waiting.
Give it a real task. For example:

Explain how this repository is organized, then suggest three small improvements.
Watch it work. The agent reads files, runs commands and edits code inside your space.
Open the inspector panel next to the terminal to see what it did:
- Activity lists the space’s audit events, such as the network requests and tool calls the agent made, and whether each was allowed.
- Requests holds questions the agent asks you.
- Blocked appears when your network policy stopped a call, with the reason and a way to approve it.
- Changes shows the files it changed.

This is the core of Marshal: the agent works for you, with only the access your organization allows, and you can see what it did.

When you finish, select Stop. With space storage set up, Marshal saves your workspace, so you can select Start on Spaces later and continue where you left off. Select Archive when you are done with a space for good. See Spaces and Spaces and storage.

Install the CLI

The CLI does the same work as the web app for spaces and runs, from your own terminal.


curl -fsSL https://marshal.codes/downloads/install.sh | sh
marshal login --server https://marshal.example.com

Replace the address with your Marshal address. A browser window opens to sign you in. You can also select Download CLI in the account menu. Full steps: Install the CLI.

Then create a space and connect to it in one command:


marshal spaces new --manifest my-manifest

What you should see: progress lines while the space starts, then a connected terminal. Every command is in the CLI reference.

Start your first run

A run is a task the agent works on alone. It ends with a result, and it asks you if it needs a decision. Hand it something you would otherwise sit and watch:


marshal run --manifest my-manifest --task "Review the repository and list problems" --follow

--follow streams the agent’s work into your terminal. Open Runs in the left rail to see the same run in the web, with its transcript and result. You can also start runs from a deputy in the web. See Start a run from the CLI, Steer a live run and Runs.

When something is blocked

What you see	What to do
A blocked domain on the Blocked tab	Select Approve for this space, or Approve & remember for next run to save it to your manifest. If neither works, the organization’s network policy denies that domain, and an admin changes the policy. See Approve a blocked domain.
A credential with Needs reconnect	Open Connections and select Reconnect.
A space in an error state	Open Home, then Needs attention, and select Start on the space.
The manifest cannot launch	Read the message on New Space. It lists what to fix, such as a missing credential.
A button is disabled with “Read-only”	Your role does not include that action. An admin can do it for you.

Home shows what waits on you, what is running, and what needs attention. A number on the Home icon counts the items that wait for you. For anything else, start at Help and fixes.

Next: hand off real work

You have a space and a run behind you. Now let an agent take a task off your plate for good: create a deputy and a mission.