🤖 MiMo SWE-Agent

Autonomous Software Engineering Agent powered by Xiaomi MiMo

---

📖 Overview

MiMo SWE-Agent is an autonomous software engineering agent that reads GitHub issues, analyzes codebases, generates fixes, runs tests, and creates pull requests — all powered by Xiaomi MiMo V2.5 Pro.

🎯 What It Does

GitHub Issue → Analyze → Plan → Code → Test → Review → Pull Request
     │           │        │       │       │        │         │
     └───────────┴────────┴───────┴───────┴────────┴─────────┘
                         All powered by MiMo

1. 📥 Read Issue — Fetches issue details from GitHub API
2. 🧠 Analyze — Understands the problem, identifies affected files
3. 📋 Plan — Creates step-by-step execution plan
4. 💻 Code — Generates precise code changes
5. 🧪 Test — Runs tests to validate changes
6. 🔍 Review — Self-reviews code quality (scored 1-10)
7. 🔀 PR — Creates pull request with detailed description

✨ Key Features

• 🤖 MiMo-Powered — Uses Xiaomi MiMo V2.5 Pro for reasoning
• 🔄 Full Pipeline — End-to-end from issue to PR
• 🌐 Web Dashboard — Real-time visualization of agent thinking
• 🎮 Demo Mode — Works without API keys (mock responses)
• ⌨️ CLI Interface — Command-line tool for quick usage
• 📊 Metrics — Token usage, success rates, timing
• 🔌 Plugin Ready — Easy to add custom agents

---

🚀 Quick Start

Installation

git clone https://github.com/Cohhuy/mimo-swe-agent.git
cd mimo-swe-agent
pip install -r requirements.txt

Demo Mode (No API Key Needed!)

# Run the full demo pipeline
python -m mimo_swe demo

# Or with CLI
python -m mimo_swe demo --verbose

Live Mode

# Set up environment
cp .env.example .env
# Edit .env with your keys

# Run on a GitHub issue
python -m mimo_swe run Cohhuy/mimo-swe-agent 1

# Check configuration
python -m mimo_swe check

Web Dashboard

# Start the web UI
python -m mimo_swe web

# Open http://localhost:8000

---

🏗️ Architecture

┌─────────────────────────────────────────────────────────────┐
│                      MiMo SWE-Agent                         │
│                                                             │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐  ┌──────────┐  │
│  │ Planner  │→ │  Coder   │→ │  Tester  │→ │ Reviewer │  │
│  │          │  │          │  │          │  │          │  │
│  │ Analyze  │  │ Generate │  │ Validate │  │ Score &  │  │
│  │ & Plan   │  │ Code     │  │ Tests    │  │ Approve  │  │
│  └──────────┘  └──────────┘  └──────────┘  └──────────┘  │
│       ↑              ↑             ↑             ↑         │
│       └──────────────┴─────────────┴─────────────┘         │
│                    MiMo V2.5 Pro                            │
│                                                             │
│  ┌──────────────────────────────────────────────────────┐  │
│  │                  GitHub Client                        │  │
│  │  Issues • Files • Branches • Pull Requests           │  │
│  └──────────────────────────────────────────────────────┘  │
└─────────────────────────────────────────────────────────────┘

Core Modules

Module	Purpose
mimo_client.py	MiMo API client (OpenAI-compatible)
github_client.py	GitHub REST API wrapper
planner.py	Issue analysis & execution planning
coder.py	Code generation from plans
tester.py	Test runner & validation
reviewer.py	Code review & scoring
agent.py	Main orchestrator pipeline
demo_engine.py	Demo mode with mock responses
cli.py	Command-line interface
web/	FastAPI web dashboard

---

🌐 Web Dashboard

The web dashboard provides real-time visualization of the agent's work:

• Dark theme with Xiaomi branding
• Live timeline showing each pipeline step
• Diff viewer with syntax highlighting
• Terminal output streaming in real-time
• Demo mode toggle for presentations

---

🎮 Demo Mode

Demo mode runs the full pipeline without any API keys:

python -m mimo_swe demo

It uses a sample Flask calculator app with intentional bugs:

• Division function uses multiplication instead
• No ZeroDivisionError handling
• Modulo endpoint not implemented
• History endpoint has undefined variable

The agent "fixes" all bugs and creates a PR — perfect for demonstrations.

---

⚙️ Configuration

Environment Variables

# Required for live mode
MIMO_API_KEY=your_mimo_api_key
GITHUB_TOKEN=your_github_token

# Optional
MIMO_MODEL=mimo-v2.5-pro          # Model to use
MIMO_BASE_URL=https://token-plan-sgp.xiaomimimo.com/v1
DEMO_MODE=false                     # Enable demo mode

CLI Options

# Run on specific issue
mimo-swe run owner/repo 42

# Run with custom model
mimo-swe run owner/repo 42 --model mimo-v2.5

# Demo mode
mimo-swe demo --verbose

# Check configuration
mimo-swe check

# Start web dashboard
mimo-swe web --port 8000

---

📊 Performance Metrics

The agent tracks detailed metrics:

• Token Usage — Total tokens consumed per step
• Success Rate — Percentage of successful PRs
• Average Duration — Mean time per issue
• Review Score — Code quality score (1-10)

📈 Execution Summary
════════════════════════════════════════════════════════
  Issue:          #42 - Calculator division bug
  Status:         ✅ PR Created
  Duration:       12.4s
  Tokens Used:    3,847
  Review Score:   9.2/10
  Tests Fixed:    4/4

🧠 Pipeline Steps
  ✅ Analyze     1.2s    450 tokens
  ✅ Plan        1.8s    680 tokens
  ✅ Code        3.4s    1,200 tokens
  ✅ Test        2.1s    320 tokens
  ✅ Review      1.9s    580 tokens
  ✅ PR          2.0s    617 tokens

---

🛣️ Roadmap

• Core agent pipeline
• Demo mode
• Web dashboard
• CLI interface
• Multi-file complex changes
• GitHub Actions integration
• Slack/Discord notifications
• Agent memory & learning
• Multi-repo support
• Custom agent plugins

---

🤝 Contributing

Contributions welcome! Please see our Contributing Guide.

1. Fork the repository
2. Create feature branch (git checkout -b feature/amazing)
3. Commit changes (git commit -m 'Add amazing feature')
4. Push to branch (git push origin feature/amazing)
5. Open Pull Request

---

📄 License

MIT License - see LICENSE for details.

---

🙏 Acknowledgments

• Xiaomi MiMo — For the powerful AI model
• SWE-Agent / Devin — For inspiration
• Open Source Community — For tools and libraries

---

Built with ❤️ using Xiaomi MiMo V2.5 Pro