Maps
A map is a scoped workspace. It owns one or more folders, a set of agents, and its own memory. Most people run a map per project.
Creating a map
Three ways to create a map, all of which hit the same code path:
- Click + New map in the titlebar tab strip.
- Press ⌘ N from anywhere in the app.
- Drag a folder from Finder or Explorer onto the empty canvas.
The first option opens a dialog, the others jump straight to the map with sensible defaults — you can rename it later.
Adding folders
A map can attach as many folders as you want. Multi-select in the file picker, or drop several from Finder in one go. Nested folders are fine: agents see them as one tree but isolation still applies at the individual folder level.
Every folder you attach becomes a permission surface for every agent on the map. Keep them narrow — a monorepo subtree is better than the whole monorepo.
Managing maps
- Rename: double-click the tab, type, enter.
- Duplicate: right-click the tab → Duplicate — copies folder attachments and agent blueprints without the chat history.
- Archive: moves the map to the archive drawer; it stops loading on launch but all history is preserved.
- Delete: shows a confirmation, then drops the map and every associated agent, channel, and message. This is permanent.
Map-level settings
Open Map settings from the tab's right-click menu. The settings apply to every new agent in the map, but existing agents keep their current config.
- Default provider: which backend new agents use — OpenAI-compatible, Ollama, or a custom endpoint.
- Default model: model string passed to the provider on each call.
- Guardrails: a list of operations that always require your approval (for example: any shell command with
rm, or any HTTP request to a new host). - Context budget: how aggressively each agent compacts its memory.