At the end of this guide, you will have an organization, a working Agent, and one small result you can inspect. The example is illustrative: an Agent drafts a Markdown launch announcement and a human reviews it. Before you begin, install and open Rudder. Have a local runtime available for the first Agent and a folder where it can save the announcement.

​
1. Name the organization

Enter a display name, such as Launch team. This first screen asks only for the name. Rudder creates the organization’s internal identity for you. You can add goals and inspect its Issue Key later from the normal work surfaces and settings.

​
2. Configure the first Agent

Choose the runtime for the first Agent and complete its environment check. A runtime is the local execution environment the Agent uses for a run. Rudder creates the Agent only after the required check succeeds, so fix any reported runtime problem before continuing. When setup finishes, Rudder opens Messenger. The Getting Started group is already marked as read, so it teaches the workflow without creating false unread work. Open its welcome item, or start the example below.

​
3. Choose Chat or an Issue

Use Chat if you want to draft and refine the announcement through one ongoing conversation. An issue is a durable task record with an explicit status and lifecycle. Use one when work needs a named owner, dependencies, or a review path; comments, agent runs, artifacts, and review decisions can stay with the same record. This guide uses an Issue because the example has an output and a reviewer. Create one with this brief:
Issue board

​
4. Make the next step executable

Move the Issue to todo only when another person or Agent could start without asking what the request means. Assign the first Agent as the single owner and add a reviewer because another person will judge the announcement. Assigning actionable work wakes the Agent automatically, so there is no separate start step. Leave work in backlog if scope, priority, or acceptance criteria still need a decision. Split independent work into separate Issues instead of assigning one Issue to two owners.

​
5. Inspect the Agent Run

Rudder invokes the configured runtime and records one Agent Run for this bounded attempt. Agent run detail When the run finishes, look for:
  • a run summary or transcript
  • the Markdown output or a link to it
  • validation or checks the Agent performed
  • a clear blocker if the result is incomplete
If there is no inspectable output, ask the Agent to attach or link it before moving to review.

​
6. Review the result

Move the Issue to in_review when the output is ready for judgment. The reviewer inspects the Issue, output, run evidence, and comments, then records a structured decision with a required comment. Approving completes the Issue automatically. Requesting changes routes it back to the assignee; a blocked decision stays visible with the reason. The review is successful when a later reader can open the same Issue and find the request, Agent Run, output, and decision without reconstructing the story from another tool.

​
Recovery

If the run fails, keep the Issue open, inspect the run evidence, and name the next action. If a missing credential or human decision blocks the task, move it to blocked and state who can unblock it. Messenger can bring that attention signal back to the responsible person.

Chat and Messenger

Decide which tasks can remain conversational.

Issue Lifecycle

Operate status, assignment, review, and close-out in detail.