START A TASK WITH SWAMP START
This guide shows you how to go from a freshly installed Swamp to an AI agent
working on your own task with swamp start, and how to come back to that task
later. swamp start runs on a desktop with a browser; for CI, containers or a
repo you set up yourself, use swamp repo init instead (see
Set up a repo directly).
Start
Run:
$ swamp startswamp start opens a page on 127.0.0.1 in your browser. The terminal keeps
the page running:
✓ Signed in to Swamp Club as alice
Open this page in your browser to continue:
http://127.0.0.1:56732/dashboard/start#token=...
This terminal keeps the page running. Press Ctrl-C to stop it.swamp start needs no credential to run. If you are not signed in, the page
shows only the sign-in step, and nothing else on the page works until you sign
in.
On a machine where no browser opens, such as over SSH, pass --no-browser and
open the printed URL yourself.
Choose a workspace
A workspace is the Swamp repo your agent works in. On the page:
- Create a workspace to get a new, empty repo under
~/.swamp/workspaces/<name>($SWAMP_HOME/workspaces/<name>whenSWAMP_HOMEis set). The page suggests a name you can change. - Import an existing Swamp repo to work in it where it is. Nothing is moved or copied.
- Pick a workspace you already have.
The same workspaces are managed from the terminal with swamp workspace; see
Workspace Commands.
Describe the task and launch an agent
- Describe the task in your own words. Swamp saves it as a brief at
tasks/<date>-<slug>.mdin the workspace. - Pick an agent. The page lists every agent Swamp supports and marks the ones
found on your
PATH; only those can be picked. It preselects the agent you picked last. You can also set the workspace up for other agents at the same time. - Confirm that you trust the workspace folder, then launch.
Swamp opens the agent in a new terminal window in the workspace, and the agent starts by reading the brief. When no terminal window can be opened, the page shows a command to copy and run yourself.
A workspace keeps every task you give it, each with its own brief, so you can add a new task to a workspace that already has one.
Come back to a task
Run swamp start again and open the workspace. Its page lists every task,
newest first, with the agent and how far it got.
Launch Agent reopens the agent without sending the task again. Claude Code, Codex and OpenCode continue their most recent conversation in the workspace; other agents open a fresh session.
To open a workspace directly, skipping the picker:
$ swamp start --workspace my-workspaceA name that is not registered yet is created under
~/.swamp/workspaces/.
More than one swamp start can have the same workspace open; the page notes
when another one has it open.
Check your workspaces from the terminal
swamp start status lists your workspaces, their tasks and the agents on your
PATH, without opening a page or changing anything:
$ swamp start status
✓ Signed in to Swamp Club as alice
Workspaces
infra-health 1 task
/Users/alice/.swamp/workspaces/infra-health
• Summarise yesterday's failed GitHub Actions runs.
created 2026-10-13 22:21 · launched with Claude Code
Agents
✓ Claude Code /Users/alice/.local/bin/claude
✓ Codex /opt/homebrew/bin/codex
✗ Cursor not on PATH (looked for cursor-agent)
...Add --workspace <name> to show one workspace. swamp start --json reports the
same information as JSON and starts nothing.
Open the dashboard
Once a workspace is open, Open dashboard in the page header shows that
workspace's behaviors, workflows, runs and data, with no swamp serve running.
A picker in its sidebar switches between workspaces. See
Local dashboard.
Stop
Press Ctrl-C in the terminal, or close the page:
Stopped. Run `swamp start` to get back to your workspaces.Set up a repo directly
swamp repo init sets up a Swamp repo in the current folder for the agents you
name, with no browser and no workspace registry. Use it for CI, containers, or
an existing project you want to set up yourself; see
per-agent setup. To add such a repo to swamp start
later, import it.