◆ Multi-Worktree

Ship in Parallel.
Stay in Control.

Run multiple agents across git worktrees simultaneously. Each agent gets its own isolated workspace, its own branch, its own context.

01 · The Problem

Context Switching Kills Flow

You're building a feature. A bug report comes in. You stash, switch branches, fix, commit, switch back, pop stash. Or worse — you wait until you're "done" before starting the fix.

With git worktrees + Glue, you just open a parallel session. Your feature work continues untouched while the bug gets fixed in its own worktree.

▸ Parallel Sessions
# Session 1: feature work (already running) ❯ glue -w feat/auth # Session 2: hotfix (new terminal) ❯ glue --new-worktree fix/crash-on-login # Session 3: refactor (another terminal) ❯ glue --new-worktree refactor/db-layer # All running simultaneously ❯ glue --list ● feat/auth claude-4-sonnet 4.5K tok ● fix/crash claude-4-haiku 1.2K tok ● refactor/db gpt-5 3.1K tok
02 · How It Works

Under the Hood

01
Isolated Checkouts
Each worktree is a full working copy with its own branch. No stashing, no conflicts. Git handles the plumbing.
02
Shared History
All worktrees share the same .git directory. Commits are immediately visible across sessions. No force-pushing needed.
03
Per-Session Context
Each Glue session maintains its own conversation history, model selection, and tool state. Fully independent.
04
Auto Cleanup
When you're done, Glue can merge, create a PR, or clean up the worktree. No orphaned directories.
05
Resource Aware
Glue monitors system resources. Won't let you spin up 20 parallel sessions on a laptop. Smart defaults.
06
Session Dashboard
See all active sessions, their status, token usage, and last activity at a glance with glue --list.
03 · Workflows

Real-World Patterns

Feature + Hotfix
Keep your feature branch running while you spin up a hotfix session. Both agents work independently, both produce PRs.
Parallel Features
Working on 3 features? Glue can run 3 sessions. Each on its own branch, its own model if you want. Review and merge when ready.
Explore & Commit
Spin up a throwaway worktree to explore an idea. If it works, merge it. If not, delete the worktree. Zero risk.
Model Comparison
Same task, different models, different worktrees. Compare Claude vs GPT on the same feature and pick the better output.