How-to Guides
New to Orbit? Start Here sets it up and walks one task
from request to merge. These guides pick up from there, one job each. Your
agent can do most of them for you with the bundled orbit-setup or
orbit-orchestrate skill.
Day to day
Section titled “Day to day”Use the Dashboard
Approve and ship tasks, follow runs, start a drain, and reach a remote host over SSH.
Use the Claude Code Mod
Act on tasks from the prompt band, inspect the board, preflight ships, and track PR runs in Claude Code.
Use the Desktop App
Supervise the dashboard server, manage the system tray, and follow fleet notifications in a native window.
Open the MCP Apps Control Center
Open the embedded control center or one task or run from an MCP client, pick a selector, and fall back when it does not render.
Run a Delivery Window
Prepare the backlog, drain it for a set time, retune or stop the window, and recover failed runs.
Schedule Recurring Work
Run jobs and file recurring chores on a schedule with the host scheduler clock, routines, and auto-tasks.
Shape the work
Section titled “Shape the work”Choose Scopes
Decide where state lives and what an activity may read or modify.
Write an Activity
Define your own agent or deterministic step in activity YAML.
Run an Experiment
Compare control and treatment arms, read intervals and the verdict, and conclude without changing configuration automatically.
Operate
Section titled “Operate”Investigate Failures from the Terminal
Trace a failure through doctor, run output, host logs, audit events, and server health checks.
Run Orbit across hosts
Register SSH hosts, route tasks by prefix, select remote workspaces, and prepare a follower.
Publish and Restore Tasks
Snapshot task records to a dedicated repository, and recover them.
Back Up and Move Orbit State
Find the state you can't regenerate, take consistent SQLite backups, verify a restore, and move tasks with export and import.
Maintain and Remove Workspaces
Inspect bindings and ship mode, converge shipped defaults, inspect migrations, rebind remotes, and remove or tear down checkouts safely.
Set Up a Distributed Drain
Spread one backlog across machines: one owner, replica checkouts, SSH access.
Report and Triage Friction
Record what made work harder than it should have been, then search, triage, resolve, and rehome the record.
Decision channels
Register a human decision channel, inspect approval digests, and connect an authenticated transport plugin.
Install and Grant Plugins
Add a plugin to this machine, grant only the permissions it requests, enable it per workspace, and upgrade or remove it.
Hold Paths with Leases
Announce or reserve the workspace paths you are editing outside a drain, then check, renew and release the lease.