Running Agents in Parallel
Create multiple panes. Each gets its own branch. Agents can’t interfere with each other.
The Workflow
Step 1 — Create panes
Press ⌘+N (macOS) or Ctrl+N (Windows/Linux) for each task. Give each pane a descriptive feature name — this becomes the branch name.
Create Pane
Creates a new git worktree with an isolated branch
Step 2 — Assign tasks
In each pane’s terminal, give the agent its instructions. Because each pane runs in an isolated worktree, agents work on independent copies of your codebase.
Step 3 — Monitor
The sidebar shows live status for every pane. No need to babysit — check in when you get notified.
runningwaitingcompletederror
Tips
- Write self-contained tasks. Agents that touch overlapping files will cause merge conflicts when you go to integrate. Scope each task to a single feature or file boundary.
- Notifications do the waiting for you. Enable desktop notifications in Settings → Notifications so you’re alerted the moment an agent finishes or needs input — no polling required.
- Review and commit each pane independently. Use ⌘+⇧+K to commit a pane’s work, then move to the next. Integrate via your normal PR or merge flow.
There is no hard limit on simultaneous panes. In practice, the bottleneck is API rate limits from your agent provider, not Pane itself.
In the New Pane dialog, set the Count field to create several panes at once. Each pane gets an auto-incremented branch name suffix (feature-1, feature-2, …). You can rename them after creation.Bulk creation is useful when you have a list of known tasks and want to queue them all before reviewing outputs.
The Dashboard panel gives a bird’s-eye view of all panes: status, branch name, last activity, and token usage at a glance. Open it with ⌘+⌥+D or from the panel switcher.From the dashboard you can archive completed panes in bulk without switching to each one individually.
Last updated on