Skip to main content

Spawning Agents

Agents are AI employees powered by Claude Code. Each agent runs independently, has its own context, and can communicate with other agents. You spawn them onto the canvas and point them at your code.

Onboarding an agent

1

Open the create dialog

Click the + button on an empty canvas cell. If the cell already has an agent, you’ll see a confirmation pill asking whether to clone it — confirm or press Escape to cancel.
2

Choose a folder

Pick which project folder the agent should work in. This becomes its primary workspace — where it reads and writes code.If your map has multiple folders, you’ll choose one as the primary. You can grant access to additional folders later.
3

Configure the agent

Optionally set:
  • Name — something descriptive (e.g., “frontend-lead”, “api-builder”)
  • Team — assign it to a team on the canvas
  • Role — categorize its specialty
4

Agent appears on canvas

The agent spawns on the spatial canvas as a node, ready to receive tasks.

Agent isolation

Each agent gets its own complete copy of the code — the same way a human engineer would clone a repo, work independently, and push when ready. When you enable Isolate new agents in settings, every agent spawned into a git repository gets:
  • A full clone of each repo it has access to — stored locally, completely independent from other agents
  • Its own branch — created automatically at spawn time
  • Remote access — origin points to your hosted remote (GitHub, etc.), so agents can push and open PRs directly
Without isolation, all agents share the same working directory. This is simpler but requires more coordination to avoid stepping on each other’s work.
Agent isolation requires git. Non-git folders are always shared regardless of the isolation setting.

Customizing an agent

After spawning, select the agent and open the Settings tab in the right panel to configure:
SettingWhat it does
NameDisplay name on the canvas and in conversations
EmojiVisual identifier for quick recognition
ColorAccent color for the agent’s node
ModelOverride the default Claude model for this agent
Folder accessControl which folders the agent can read or write

Deleting an agent

Select the agent and press Cmd+R to terminate it, or right-click and choose “Delete” from the context menu.
Deleting an agent removes its conversation history and settings. If the agent was working on a branch, the branch remains in your repo — only the Pentagon agent is removed.

Next: Agent Identity

Give agents persistent personality, purpose, and memory. Context compounds.