This guide takes you from a new installation to one completed Agent task. You will create an organization, connect the runtime you already use, and ask for a small result in Chat. Before you begin, install and open Rudder. If you plan to use a local runtime such as Codex or Claude Code, install it and sign in first.

​
1. Create your organization

Rudder calls a team workspace an organization. Enter a name you will recognize, such as your project, company, or personal workspace name. That is enough to start. You can add goals, projects, and more Agents later.

​
2. Connect your first Agent

Choose the runtime already available on your computer, then follow the setup form:
  • Choose Codex or Claude Code if you use its local CLI.
  • Choose a local command for your own script or Agent process.
  • Choose a webhook or remote option only when you already have that service.
Run the environment check. A successful check means Rudder can start the runtime; it does not spend time on a real task yet. If the check fails, use the message in the form to install, sign in, or correct the command.

​
3. Ask for one useful result

Open Chat and use a request small enough to judge in a few minutes. For example:
A good first request says who the result is for, what you want back, and any limits that matter. Avoid “test Rudder”—it gives you nothing useful to judge.

​
4. Follow the run

Rudder starts the selected Agent and shows its reply in Chat. You can continue using the same conversation to clarify the request, stop an unhelpful run, or send a follow-up while it is working. Open the Agent Run when you want more detail. It shows which runtime handled the request, whether the run finished, and the available transcript, usage, errors, or output.

​
5. Judge the result

Read the actual answer or open the file the Agent produced. Then choose the lightest next step:
  • Ask for a revision in the same Chat.
  • Save or share the result if it is ready.
  • Create an Issue if the task now needs an owner, dependencies, a visible status, or an independent reviewer.
You have completed the first task when you can find the original request, the Agent’s result, and the run that produced it from the same conversation.

​
If the Agent cannot finish

Open the failed Agent Run and read the plain-language error first. Common causes are a runtime that is not signed in, a missing model or command, a folder the runtime cannot access, or a permission the task needs. Fix the named problem and retry the small request. Do not add more Agents, projects, or automation until one simple Chat task works.

Chat or Issue?

Learn when a conversation is enough and when to add structure.

Configure a runtime

Fix or fine-tune the environment an Agent uses.