Zashiki Warashi

vs lpm

lpm and Zashiki both care about multiple local projects on a Mac. The fork in the road is configuration tax versus observe-only catalog — and whether AI agent terminals are in scope.

Short answer

lpm (and similar local project managers) shine when you want a defined multi-service workspace — often with explicit config, live terminals, and AI-agent-oriented flows.

Zashiki Warashi shines when you want a house spirit for folders you already have: scan roots, infer start commands, stop process groups, peek Compose DBs, open Compass — without taxing every AI-generated repo with a new schema.

Side-by-side

Concern lpm (typical) Zashiki Warashi
Project model Services you define Observe existing repo signals
Per-project config Central to the workflow Optional later; none required in v1
AI coding agents in-app Strong focus in current lpm Out of scope
Compose DB URI / Compass Not the core story First-class peek
Port reconciliation Varies Catalog / docker / native occupants
Cloud accounts No (local) No (local)

Exact lpm features move quickly — treat their site as source of truth. This page is about category fit, not a scorecard of version N.

Who should pick Zashiki

  • You have a pile of Cursor / vibe-coded folders and refuse another YAML per folder
  • You care about Compose DB peek and Compass more than embedded agent terminals
  • You want login-shell start/stop with process-group semantics — Start and stop

Who should pick lpm (or similar)

  • You want one workspace that runs services and AI agent terminals side by side
  • You are happy maintaining explicit service definitions and profiles
  • Duplicating checkouts for parallel agents is a primary workflow

Honest product principle: prove the mechanism; do not invent superiority. Both can be right for different Mondays.

Download Zashiki: Releases.

FAQ

Which one needs YAML in the project?

lpm is built around defining services (and profiles) in config. Zashiki infers start commands and keeps overrides in app data — no required per-repo YAML.

Does Zashiki run Claude Code or Codex terminals?

No. It is not an agent workspace. It operates the projects agents leave behind and can open Cursor.

Can I use both?

Yes in principle — they optimize different workflows. Pick one as the daily house to avoid duplicate lifecycle ownership.

Download for macOS GitHub