Mozzie
FreemiumCodex Claude Gemini CLI parallel agents orchestration
What does this tool do?
Mozzie is a local-first desktop application that orchestrates multiple AI coding agents (Claude Code, Gemini CLI, Codex) to work in parallel on software development tasks. Users describe what needs building in natural language, and an LLM orchestrator breaks it into discrete work items with dependencies, assigns them to agents, and manages isolated git worktrees for each task. The system handles the complete workflow: agent execution with live output streaming, dependency tracking with automatic launching of blocked tasks, code review with diff visualization, and feedback injection for failed attempts. Everything runs locally on the user's machine—no cloud dependency—with a single-window interface for managing parallel agent work, reviewing diffs, and merging approved branches.
AI analysis from Mar 13, 2026
Key Features
- LLM orchestrator (OpenAI, Anthropic, Gemini) that decomposes natural language requirements into work items with automatic dependency graphs and agent assignment
- Git worktree isolation with branch management per work item, preventing agent conflicts and enabling parallel independent execution
- Live streaming agent output with tool-call activity visualization and terminal replay for debugging
- Multi-agent parallel execution supporting Claude Code (ACP protocol), Gemini CLI, Codex CLI, and custom CLI tools
- Feedback loop system that injects rejection reasons and attempt history into agent prompts on re-run
- Dependency tracking with cycle detection and cascading auto-launch of blocked work items
- Sub-work-items (stacked branches) where children merge into parents before parent pushes to origin as single PR
- Review workflow with diff visualization, approve-to-merge, and reject-with-feedback state transitions
Use Cases
- 1Solo developers or small teams who want a personal build team running on their local machine without cloud costs
- 2Projects requiring parallel feature development where multiple independent tasks can be executed simultaneously by different agents
- 3Development teams practicing code review workflows where rejected work gets automatic feedback loops to prevent repeated mistakes
- 4Complex codebases with dependency-heavy tasks where the orchestrator tracks which work items block others and launches them automatically
- 5Developers managing multiple projects simultaneously who need a unified interface to coordinate agents across workspaces
- 6Organizations building with stacked branches and child-to-parent merge workflows that benefit from sub-work-item hierarchies
Pros & Cons
Advantages
- True local-first architecture with SQLite and git worktrees means zero cloud dependency, full offline capability, and complete data privacy—no agent execution leaves your machine
- Intelligent feedback loops inject rejection reasons and attempt history into agent prompts, enabling agents to learn from failures and avoid repeating mistakes across sessions
- Dependency graph management with cycle detection and cascading auto-launch eliminates manual coordination bottlenecks—blocked work items launch automatically when dependencies complete
- Multi-agent parallel execution with agent-agnostic design (supports Claude Code ACP, Gemini CLI, Codex, or custom scripts) maximizes hardware utilization
- Persistent orchestrator conversation history carries context across sessions, allowing users to pick up interrupted work without re-explaining context
Limitations
- High setup complexity with multiple prerequisites (Node 20+, pnpm, Rust, Tauri dependencies, agent CLIs) creates friction for non-technical users or teams with diverse dev environments
- Completely local execution means scalability is hardware-bound—users limited by their machine's CPU/memory; no ability to offload to more powerful infrastructure
- Early-stage project with actively developing features and likely stability issues; production readiness unclear and documentation appears incomplete (content cuts off mid-sentence in Features section)
- Requires manual configuration of multiple API keys and agent settings before use, with no clear migration path for teams already using cloud-based alternatives
- Limited agent ecosystem—while extensible, real-world support appears limited to Claude Code (full ACP support) with other agents via CLI fallback, potentially reducing reliability
Pricing Details
Pricing details not publicly available. Mozzie appears to be open-source software available on GitHub. Users incur costs only for LLM API calls (OpenAI, Anthropic, or Gemini APIs configured in settings) based on their chosen orchestrator provider's pricing.
Who is this for?
Solo developers and small engineering teams (2-10 people) who want local control over AI-assisted development without cloud vendor lock-in. Best suited for technically sophisticated users comfortable with Rust/Node development environments, git workflows, and API key management. Ideal for teams practicing code review discipline, managing complex multi-part features, or experimenting with AI agents for code generation.