🚀 Fleet Agent IDE

The Open-Source Multi-Agent Orchestrator with Git Worktree Isolation & Self-Healing Loops

Run fleets of autonomous AI coding agents concurrently on your local repository with zero file collisions, zero git locks, and zero Docker latency.

Quickstart • Superpowers • Web Dashboard • Architecture • Documentation

---

⚡ The Problem: Why Traditional AI Coding Tools Get Stuck

Most developer AI tools modify your active branch directly or run inside heavy virtual containers:

• 🛑 You can only run 1 task at a time — your editor gets locked while the agent works.
• 💥 Multi-agent execution causes merge chaos — agents overwrite each other's files and trigger .git/index.lock errors.
• 🐢 Docker/VMs are slow and resource-heavy — wasting 10GB of RAM and taking 15s to start.

---

💡 The Fleet Solution: Native Git Worktrees + LangGraph

Fleet Agent IDE replaces heavy containers with native Git Worktree Isolation:

graph TD
    User["👨‍💻 Developer (Web Dashboard / CLI)"] <-->|REST + SSE Stream| Daemon["⚡ Fleet Backend Daemon (FastAPI)"]
    
    subgraph Core Engine
        Daemon <--> State["💾 Async SQLite Store (contracts/)"]
        Daemon --> Scheduler["⚙️ Fleet Scheduler & DAG Resolver"]
        Scheduler --> Planner["🧠 Supervisor / Planner Agent"]
        Scheduler --> WorktreeMgr["🛡️ Git Worktree Isolation Engine"]
    end
    
    subgraph Isolated Worktrees & Parallel Agent Fleet
        WorktreeMgr -->|Allocate agent/sub-1| WT1["📁 Worktree 1 (.fleet/worktrees/sub-1)"]
        WorktreeMgr -->|Allocate agent/sub-2| WT2["📁 Worktree 2 (.fleet/worktrees/sub-2)"]
        
        WT1 --> LangGraphWorker1["🤖 LangGraph Worker (Reflective Loop)"]
        WT2 --> LangGraphWorker2["🤖 LangGraph Worker (Reflective Loop)"]
        
        LangGraphWorker1 --> Test1{"🧪 Run Local Tests"}
        Test1 -- "❌ Exit != 0 (stderr)" --> Fix1["🔁 Self-Healing Reflection"]
        Fix1 --> LangGraphWorker1
        Test1 -- "✅ Exit == 0" --> Ready1["🎉 Ready to Merge"]
    end

Loading

1. Instant Spawning (<50ms): Spawns independent physical working copies on dedicated branches (agent/<task-id>).
2. Autonomous Self-Healing Loop: The agent runs your local tests (pytest, npm test, cargo test), intercepts stderr, and fixes bugs iteratively before notifying you.
3. Works 100% Offline with Ollama ($0 Tokens) or Free Gemini 2.0 Flash: Zero required paid subscriptions.
4. Live Web Dashboard & Diff Viewer: Track agents in a multi-column Kanban board and inspect side-by-side Monaco diffs before merging.

---

🖥️ Web Dashboard

The Fleet Web Dashboard (web/) provides a visual control room for your agent fleet:

• 📊 Multi-Agent Kanban Board: Real-time state columns (Planning, Working, Self-Correction Loop, Ready to Merge).
• 🔍 Monaco Diff Inspector: Side-by-side git diff viewer with syntax highlighting and 1-click [Merge to Main].
• 📺 Live SSE Terminal Streamer: Watch agents execute tests and see stdout/stderr streamed live.

---

🚀 Quickstart in 3 Minutes

1. Clone and Install

git clone https://github.com/DrowLink/fleet-agent-ide.git
cd fleet-agent-ide

# Create & activate environment
python -m venv .venv
# On Windows: .venv\Scripts\Activate.ps1
# On Linux/macOS: source .venv/bin/activate

# Install packages in editable mode
pip install -e "./contracts"
pip install -e "./backend"
pip install -e "./cli"

2. Configure Your LLM (Free Gemini or Offline Ollama)

Create a .env file in the root:

# Option A: 100% Free Gemini Flash (1,500 daily requests via Google AI Studio)
LLM_PROVIDER="google"
GOOGLE_API_KEY="your-google-ai-studio-key"

# Option B: 100% Offline Local via Ollama ($0 tokens)
# LLM_PROVIDER="ollama"
# OLLAMA_MODEL="qwen2.5-coder:7b"

3. Launch the Daemon & Web Dashboard

# Terminal 1: Start Backend Daemon
fleet daemon --port 8000

# Terminal 2: Start Web UI
cd web
npm install
npm run dev

Open http://localhost:5173 in your browser!

---

💻 CLI Usage (Terminal-First Developers)

# Dispatch a new task to the fleet
fleet run "Refactor authentication middleware to use JWT" --title "JWT Auth"

# Inspect task lifecycle statuses
fleet status

# View currently active Git worktrees
fleet worktrees

---

📂 Monorepo Architecture

fleet-agent-ide/
├── contracts/                         # 📜 Typed Pydantic schemas (Tasks, Events, Harnesses)
├── backend/                           # 🧠 FastAPI Daemon, Worktree Manager & LangGraph
├── cli/                               # 💻 Typer + Rich Terminal Interface
├── web/                               # 🖥️ React + Vite + Tailwind + Monaco Dashboard
├── docs/                              # 📚 Architecture Decision Records (ADRs) & Guides
│   ├── adr/                           # ADR 0001 (Worktrees), ADR 0002 (SSE), ADR 0003 (Loops)
│   └── superpowers/                   # Deep dives into core capabilities
├── AGENTS.md                          # 🤖 Operational context for AI coding agents
├── CLAUDE.md                          # 📋 Claude Code conventions & commands
├── CONTEXT.md                         # 🌐 Deep domain and distributed system context
├── DESIGN.md                          # 🎨 UI/UX design tokens and component specs
├── CONTRIBUTING.md                    # 🤝 Contributor guidelines & PR etiquette
└── LICENSE                            # 📄 MIT License

---

🗺️ Roadmap

• Git Worktree Isolation Engine
• LangGraph Self-Correction Feedback Loop
• FastAPI Daemon with REST + SSE Streaming
• Interactive Terminal CLI (fleet run, fleet status)
• React + Vite + Monaco Web Dashboard
• Ollama Local Offline & Google AI Studio Free Tier Support
• Tree-Sitter Repository Map Generator (Aider-style AST indexing)
• Surgical Search/Replace Patch Engine
• Automated Multi-Branch Merge Queue with Conflict Resolution Agent

---

🤝 Contributing

We welcome contributions of all kinds! Please check out CONTRIBUTING.md to get started.

📄 License

Distributed under the MIT License. See LICENSE for more information.