Worktrees

Git worktrees let you run multiple Glue sessions in parallel, each in its own isolated working copy. No stashing, no conflicts between sessions.

How It Works

A git worktree is a full checkout of your repository on a separate branch. All worktrees share the same .git directory, so commits made in one worktree are immediately visible across all of them.

1. Isolated checkouts — each worktree has its own branch, so there is no need to stash or switch branches
2. Shared history — all worktrees reference the same .git directory; commits, tags, and refs stay in sync
3. Per-session context — each Glue session has its own conversation history and tool state, including a record of which worktree it was started in

Worktree creation, cleanup, merging, and PR opening are managed with git worktree and your usual git tooling — Glue itself doesn't create or destroy worktrees, it just runs inside whichever one you launch it from.

Setup

Create worktrees with git, then launch a Glue session in each:

# Create worktrees with git
git worktree add worktree/feat/auth feat/auth
git worktree add worktree/fix/crash fix/crash
git worktree add worktree/refactor/db refactor/db

# Launch a Glue session in each
cd worktree/feat/auth && glue
cd worktree/fix/crash && glue
cd worktree/refactor/db && glue

Each session operates independently with its own branch, file state, and conversation.

Common Workflows

Feature + Hotfix

Keep feature work running while fixing a bug in a separate worktree. The hotfix can be merged to main without touching your in-progress feature branch.

Parallel Features

Run 3 sessions on 3 features simultaneously. Each agent works on its own branch with no interference from the others.

Explore & Commit

Create a throwaway worktree to experiment freely. If the experiment works, merge it. If not, remove it with git worktree remove <path>.

Model Comparison

Run the same task with different models in different worktrees. Compare the results side by side to evaluate quality, speed, or cost.

TIP

Worktrees are especially useful for long-running agent tasks. You can start a feature in one worktree and continue working manually or with another agent session in a different worktree.