Mozzie

Local-first desktop app that orchestrates AI coding agents in parallel

Mozzie takes a different angle from other parallel agent tools: instead of just giving agents isolated worktrees, it acts as an intelligent project manager. You describe what needs building, and Mozzie uses an LLM to break the work into items, assign agents, track dependencies, and queue everything for your review — all running locally on your machine. Built on Tauri 2.0 with a Rust backend and React frontend, Mozzie integrates with Claude Code, Gemini CLI, Codex, and custom scripts. The dependency graph with cycle detection auto-launches tasks in the correct order. When you reject a work item, Mozzie injects your rejection reason plus the full attempt history into the agent's next prompt — so it doesn't make the same mistake twice. API keys stay in the OS keychain and never leave your machine; the only external network call is the orchestrator LLM call when you use the command bar. At v0.1.8 and 44 stars, this is very early but represents a thoughtful architecture for teams who want agent orchestration without cloud lock-in.

Panel Reviews

Ship

“The rejection feedback loop is the killer feature here — most orchestration tools just retry blindly. Injecting the full attempt history plus your reason into the next prompt is the kind of detail that separates tools built by engineers who've felt the pain. Early but worth watching.”

Skip

“44 stars and v0.1.8 means you're signing up to debug someone else's early-stage orchestration system. The dependency graph sounds great until you hit a gnarly merge conflict that the LLM-based planner didn't anticipate. The local-first angle is a selling point but also means no shared state for teams.”

Skip

“Intelligent task decomposition with agent assignment and dependency tracking is what post-IDE development looks like. Mozzie is building the project management layer that every agentic coding workflow will eventually need.”

Skip

“The onboarding friction for a 0.1.x Tauri app is real. Once this hits 1.0 with better docs I'd revisit — the privacy-first design is exactly what I want for client projects.”