Each project's detail panel lists every agent the backend says is in the project, with a per-agent enable toggle. The toggles are local to this machine.
What "enabled" does
When an agent is enabled in a project on this Build Station:
- Build Station opens an SSE listener for that agent (so dashboard task pushes arrive in real time).
- The queue poller fetches that agent's queued tasks for that project on every tick.
- Tasks for that agent in that project spawn here.
When it's disabled:
- No SSE listener; no poll fetches.
- Tasks for that agent sit on the server queue (or get picked up by a different Build Station you've signed in elsewhere).
Note: Platform agents auto-enable
Agents marked as Platform on the dashboard (those with "has access to all projects") get auto-enabled the first time a project is synced. If you don't want them running on a specific machine, just toggle them off — your choice sticks across resyncs. See Platform agents vs project agents.
When to disable agents
A few legit reasons:
- You only want one agent running here. E.g. you've enabled developer + infra + reviewer at the workspace level but on this laptop you only want the developer agent. Disable the other two.
- Resource constraints. Each agent spawn is a full Claude Code session — that's RAM, CPU, and tokens. On a small machine, fewer concurrent agents = better.
- You don't trust a specific role on this checkout. Maybe you've got production credentials in your environment and don't want the infra agent doing anything autonomous. Disable it; the dashboard still works, it just won't act from this machine.
When NOT to disable
- Just to "pause" work temporarily. Toggle the project off instead — that pauses everything, comes back the same as before when re-enabled. Per-agent toggles are meant to express a steady-state preference.
- To "delete" an agent. Agents are server-side. Disabling here removes nothing on the backend; the agent still exists and someone else's Build Station can still run its tasks. To actually delete an agent, do that on the dashboard.
Multiple agents on one project
You can absolutely have several agents enabled on a single project. Common shape: developer + infra + reviewer, each handling tasks for its role.
Build Station serializes their work — at most one Claude session runs at a time per project working directory, regardless of how many agents are enabled. The others queue and pick up automatically when the running one finishes. See One task at a time per project.
If you want true parallel work (multiple agents in the same project simultaneously), enable worktree mode — each spawn gets its own isolated git worktree.
The dashboard agent list vs the local list
The list of agents in the project detail panel comes from the backend. If you don't see an agent you expect:
1. Resync
Click Resync on the project. Build Station re-fetches the agent list. New agents added on the dashboard show up; agents removed disappear.
2. Check the dashboard
Go to the project on the dashboard and look at its agent assignment. If the agent isn't there server-side, Build Station won't show it either.
3. Check Platform status
If the agent is a Platform agent, make sure it hasn't been excluded from this project server-side.