Spawning Agents

How to create and manage agents in your workspace.

From the dashboard

Click New Agent in the dashboard title bar. The spawn dialog asks for:

FieldDescription
NameA human-readable identifier (e.g., api-refactor). Other agents can message by name.
TaskA natural language description of what the agent should do. Be specific — this becomes the agent's primary instruction.
Serviceclaude or codex — which AI to use.

The working directory defaults to the project you opened (via Open Project), or you can set it per-agent.

From another agent

Any running agent can spawn new agents to delegate work. For example, one agent discovers a problem and spawns a specialist to fix it. The spawning agent provides a name, task, and optionally the service type and working directory.

What happens at spawn

  1. A git checkpoint is taken in the working directory (if it's a git repo), so you can roll back the agent's changes later
  2. Clustr configures the agent with collaboration tools and instructions
  3. The AI is launched in its own terminal with your task
  4. Clustr handles all startup prompts and permissions automatically
  5. The agent appears in the dashboard and can start collaborating with other agents

Managing agents

From the dashboard, you can manage any agent:

  • Kill — stops the agent but keeps its record so you can review what it did
  • Remove — stops the agent and deletes its record entirely
  • View diff — see exactly what files the agent changed since it started
  • Rollback — undo all of the agent's changes and restore your code to how it was before the agent started