Worktrees are the core feature of Termpad—isolated workspaces where you run AI agents in parallel. Each worktree has its own copy of your codebase, terminal, and branch, so multiple AI agents can work simultaneously without interfering with each other.Documentation Index
Fetch the complete documentation index at: https://docs.termpad.com/llms.txt
Use this file to discover all available pages before exploring further.
Why Worktrees Matter
Traditional AI coding workflows are sequential: give an AI a task, wait for it to finish, review, then start the next. This wastes time. With worktrees, while one AI builds a feature, you can have others tackling different tasks in parallel.| Component | Isolation |
|---|---|
| Git worktree | Each worktree has its own copy of the codebase |
| Terminal | Separate terminal instance per worktree |
| Branch | Each worktree works on its own branch |
| Ports | Configurable port assignments to avoid conflicts |
Worktrees share the same git history, but file changes are isolated until you merge.
Prerequisites
- A project added to Termpad
- Understanding of what you want the worktree to accomplish
Creating a New Worktree
Enter a branch name
Type a descriptive branch name. Good naming conventions:
feature/login-pagefix/null-pointer-errorrefactor/api-clientexperiment/new-caching
Choose a base branch
By default, new branches are based on
origin/main. Select a different base if needed.Worktree Anatomy
Each worktree includes:| Component | Description |
|---|---|
| Git worktree | Isolated copy of your codebase on a separate branch |
| Main terminal | Where you run your AI agent (Claude, Gemini, etc.) |
| User terminals | Additional terminals for dev servers, tests, etc. |
| File changes | Shows uncommitted changes in this worktree |
| Diff viewer | Review changes before committing |
Using an Existing Worktree
If you want to import worktrees that already exist:Worktree Workflow
A typical worktree workflow:Managing Multiple Worktrees
Create additional worktrees to work in parallel:Deleting a Worktree
When a worktree’s work is complete:Best Practices
Choose independent tasks
For best results, choose tasks that don’t overlap:- Good
- Avoid
- Worktree 1: Build login page
- Worktree 2: Build settings page
- Worktree 3: Add API rate limiting
One task per worktree
Keep worktrees focused on a single task. This makes:- Progress easier to track
- Changes easier to review
Use descriptive branch names
Your branch name should describe the work:| Type | Example |
|---|---|
| Feature | feature/user-authentication |
| Bug fix | fix/login-redirect-loop |
| Refactor | refactor/extract-api-client |
| Experiment | experiment/graphql-migration |
Clean up finished worktrees
Don’t let old worktrees accumulate. Delete them when:- The PR is merged
- You’ve decided not to pursue the work
- The experiment is complete