Get started
Your first run in the console
Sign in at cloud.getrush.ai, connect GitHub, create a Project and start a run, then watch it work and open the pull request it made.
The console at cloud.getrush.ai is where you start and steer work by
hand. It calls the same API your code does, so a run you start here is the same kind of session a
POST /o/{org}/sessions starts. This page takes you from signing in to a pull request ready for your review.
Sign in
Open cloud.getrush.ai and press Continue with Phoenix ID. A Phoenix ID window opens and the page
shows Approve the code with a short code; approve it there and the console opens on Home.
The same Phoenix ID signs you in to the rush CLI.
The sidebar has Home, Runs, Agents, Projects and Settings.
Connect GitHub
A run that works on a repository reaches it through the Rush GitHub App. Open Settings → Integrations, find GitHub under Repositories and apps, and press Connect. GitHub asks which account or organization to install on and which repositories to allow. Once installed, the button reads Add account or org for the next one.
A run with no repository needs no GitHub at all; skip to the composer below with No project selected.
A run also needs a model account. Under Settings → Accounts, connect a Claude or ChatGPT subscription or an API key. With none connected, a run uses the one-time $1 of models Rush covers for a first try, which the composer marks Models on us.
Create a Project
A Project holds a repository, its defaults and its runs, and everyone in your org can see it. Open Projects and press New project:
| Field | What to enter |
|---|---|
| Repository | One the GitHub App can reach |
| Project name | Fills in from the repository |
| Default branch | Repository default, or Change to pick one |
| Linear project (optional) | A Linear project, if you connected Linear |
| Default harness | The agent its runs use unless you pick another |
Press Create project and the console opens the Project's page.
Start a run
The composer sits at the bottom of every console page. On a Project's page the Project is already filled in. Pick a Harness (Claude Code, Codex or OpenCode), and optionally a model, an effort and a saved agent. Then write the task the way you would brief a colleague:
The checkout test fails about one run in five. Find out why, fix it, and open a PR.
Press Enter to start it (Shift+Enter adds a line). The composer shows Starting your run… and then opens the run's page.
Watch it work
The run page has three parts:
- The transcript, in the middle: what the agent says and each command and edit it makes, as it happens. The header shows the run's status and the repository, branch and short id.
- Live display, on the side: the run's own computer and browser, read-only while you watch. As the person who started the run, or an org admin, press Take control to drive it for five minutes at a time; Release control hands it back.
- The dock, below it: Environment, Terminal and Files. Files shows the Changes the agent made, the files you Attached, its Artifacts and the Workspace.
The composer becomes Follow up on this run. Send a message mid-turn and it waits in Queued messages until the turn ends; Stop ends the current turn so a queued message goes through sooner.
Open the pull request
When the work is code, the run's proof is a pull request with its checks. The header links it as
PR #412, with its number; open it to review the diff on GitHub. If the checks did not pass, the run ends
needs_review instead of completed, and a follow-up such as "the lint job fails, fix it" continues
in the same sandbox on the same branch.
Runs lists every run in your org, with who started it and its status.
Next
- The CLI in two minutes starts the same runs from a terminal.
- Watch and steer covers follow-ups and the live display in depth.
- Your first run from code starts one with an API key.