Snowtree is Databend Labs' review-driven workflow for keeping AI coding sessions safe, auditable, and merge-ready.
AI generates code. You must review. You can't review everything or roll back safely.
Snowtree fixes this with worktree isolation, incremental review, and staging snapshots.
- Worktree isolation – every AI session runs in its own Git worktree, so you can spike multiple ideas in parallel with zero merge headaches.
- Incremental review loop – review, stage, and lock in vetted changes after each AI round; subsequent rounds only diff against staged code.
- Native CLI agents – run Claude Code or Codex directly without wrappers, meaning no extra queues or limits.
- Stage-as-snapshot – staged files become the canonical baseline. When you're ready, merge them back and ship the PR.
- AI agent writes code – edits live in the isolated worktree while you review.
- AI agent commits – generates messages and commits the staged snapshot.
- AI agent syncs PRs – opens or refreshes pull requests on demand.
- AI agent updates from
main– rebases/merges the latest upstream changes. - AI agent resolves conflicts – fixes merge conflicts without touching staged files.
Install at least one AI coding agent:
| Agent | Install |
|---|---|
| Claude Code | npm install -g @anthropic-ai/claude-code |
| Codex | npm install -g @openai/codex |
One-line installer (macOS/Linux):
curl -fsSL https://raw.githubusercontent.com/databendlabs/snowtree/main/install.sh | shManual download: GitHub Releases
| Platform | Format |
|---|---|
| macOS | .dmg (arm64, x64) |
| Linux | .deb, .AppImage (x86_64) |
make install # Install dependencies
make run # Start development server
make check # Typecheck, lint, and test
make build # Build packagesSnowtree: Review-Driven Safe AI Coding
Apache-2.0