Runs multiple AI coding agents on the same repository in parallel, each in its own git worktree with its own model, then reconciles their changes with explainable, ordered merge rules instead of another model guessing.

Why#

Three agents on one checkout means trampled files and lost work. worktree-orchestrator gives each task its own isolated git worktree, created and cleaned up automatically, so agents stop overwriting each other.

How it works#

A CLI drives the whole flow: plan to define tasks, run to execute them in parallel, status to monitor, merge to reconcile, clean to prune worktrees. Merging uses a neutral reconciler: clean, automatic merges for non-overlapping paths, and a review queue for anything that actually overlaps, rather than asking another LLM to guess at intent. Each task can also route to a different model, so cheaper models handle simple tasks while harder ones get a stronger model.

Status#

v0.x with the core plan, run, merge and clean flow working. A live conflict watcher and a TUI dashboard are in active development.