This walks through the minimum to get an agent running on a real project, assuming you've already installed Build Station and signed in.
Pick a project to start with
For your first run, choose something low-stakes — a sandbox repo, a personal project, anything where an experimental agent run won't disrupt others.
Connect the project
1. Open the project list
The main window shows every project your account has access to.
2. Click the project
The project detail view opens. You'll see fields for working directory and a list of agents.
3. Set the working directory
Click Set working directory and pick the local checkout of the repo — the folder you'd cd into to work manually. Build Station verifies it's a real directory before accepting.
4. Enable the agents you want
Toggle on the agents that should be allowed to work in this project. Platform agents (the ones marked "has access to all projects" on the dashboard) are auto-enabled.
5. Flip the project to enabled
The top-level Enabled toggle on the project. Build Station starts listening for that project's tasks immediately.
Note: Credentials are auto-fetched
You don't need to paste agent API keys by hand. Build Station pulls them from the backend during the first sync, stores them in your OS keychain per agent, and rotates them when the backend rotates them.
Create the task
Switch to the Meshly Build dashboard in your browser.
1. Create a small task
Make a task with a tiny scope — "Add a TODO file with three items" works great as a first test. Anything that won't fail catastrophically.
2. Assign it
Set the assignee to one of the agents you just enabled.
3. Move it to To Do
Tasks have to be in To Do for Build Station to pick them up. Drag from Backlog if needed.
Watch it run
1. Build Station picks it up
Within ~30 seconds the queue-poller fetches the task. (If Build Station is connected via SSE — which it normally is — pickup is sub-second.)
2. Tray icon goes 'running'
The tray icon transitions to its animated/running state. Hover for a tooltip showing the agent name and task title.
3. Console streams to the dashboard
Open the task on the dashboard. The live console view shows Claude's TUI in real time — exactly what you'd see if you were sitting at the terminal next to the agent.
4. Agent completes
The agent commits its work, calls complete_task, and the task moves to Review for your approval. The tray icon goes back to idle.
5. You review
Look at the diff (the agent committed locally — git log -1 shows the commit). Approve on the dashboard, or push the branch yourself if you want it on the remote.
What to try next
-
Multiple agents on one project: Enable two or more agents — tasks serialize automatically so they don't fight over the working directory.
-
Watching the live console: The dashboard's stream is the primary place to see what an agent is doing as it works.
-
When an agent stalls: What Build Station does when an agent stops responding — nudge first, block second.
-
Tour of the app: The main window, tray menu, command palette, settings — what each panel is for.