Your first agent run

By the end of this tutorial you will have launched an AI coding agent against a project, watched it work in live mode, and read back exactly what the run cost. It takes about ten minutes and assumes you can already sign in to a SupaCloud workspace with at least one project.

What you need

• A SupaCloud account in a workspace (admin@example.com if you just self-hosted).
• One project in that workspace. Any git project works.

1. Open the cockpit

Sign in and you land on the dashboard — the operator cockpit. It shows your active and waiting runs and a live run stream, plus panels for the runner fleet, spend, and an attention queue. Every run you launch surfaces here.

New workspace? The dashboard guides you

On a brand-new workspace the dashboard greets you with a short Get started checklist — connect a model harness, add Git access, create a project, then launch your first run, credentials-first in that order. The matching harness login is used automatically (no wiring step), and optional MCP and skill steps stay out of the way. Finishing it graduates the workspace, lifting the onboarding autonomy cap so the Autonomous Delivery Engine can start dispatching on its own. This tutorial covers the same path in detail.

2. Launch a run from the web terminal

SupaCloud ships a web terminal at /cli that drives the same service layer as the UI. It is the quickest way to launch your first run.

1. Open /cli from the sidebar (or navigate to it directly).

2. List your projects to confirm the name you will target:

   projects

3. Launch a run. The run command takes a project, an agent, and a prompt, then drops you straight into live mode:

   run my-project claude "Add a /health endpoint that returns 200 and the build SHA"

   • my-project — the project name from step 2.
   • claude — the agent to use.
   • everything after that — the prompt for the agent.

Start from an issue instead

If the work already lives in a tracker issue, seed the prompt from it with run my-project claude --issue 42. See Launch a task from an issue.

3. Follow the run live

After run, the terminal is in live mode: the agent’s thinking, tool calls, and messages stream in as they happen. In live mode you don’t need a command prefix — just type and press Enter to send the agent an intervention (a nudge, a correction, an answer to a question). Press Ctrl+C to leave live mode without stopping the run.

If you ever leave and want to come back, re-enter live mode with the run’s id:

follow <id>

You can also intervene from outside live mode with intervene <id> <message>, or stop a run with cancel <id>.

4. Read the result and its cost

When the agent finishes, open its run detail page — click the run from the dashboard or the runs list to peek at it, then click Open task (or Open workflow) in the drawer header to open its full detail page (each run is also a deep-linkable URL). The panel shows the trigger, the duration, the start and finish timestamps, the container the run used, and — right in the header metarow — the agent cost in USD. You don’t need to leave the UI to see it.

The same figures — status, cost, duration and last activity — are also available from the web terminal, if you prefer the keyboard:

task <id>

What you learned

• The dashboard is the cockpit for every run.
• run <project> <agent> <prompt> launches a task and enters live mode.
• In live mode, Enter sends an intervention; Ctrl+C exits.
• The run/task detail panel shows status, duration, timestamps and cost right in the UI; the web terminal’s task <id> reports the same.

Next steps

• Launch a task from an issue
• Web terminal command reference
• Workspaces and Organizations