🧠 commit-intel

LLM-powered local Git commit review CLI tool for Android engineers.

commit-intel runs a local AI agent pipeline using Ollama and LangGraph to review staged or committed code diffs, providing feedback on:

• 🔍 Summary
• 🧪 Code critique
• 💡 Suggestions
• 🔐 Security issues
• 📐 Architecture
• 🧪 Test coverage
• 🖼️ UI changes (Jetpack Compose or XML)
• 📦 Dependency usage
• ⚡ Performance
• 🔤 Readability

---

✅ Features (Phase 1–2 Complete)

✅ Local-first

• Uses Ollama and local models like codellama:7b (no cloud API calls required)

✅ CLI Tool

• Run via simple terminal commands:

  review commit
  review commit --diff HEAD~2
  review staged

✅ Git Pre-commit Hook

• Optional hook setup:

  review install-prehook

✅ Multi-agent LangGraph Review Pipeline

• Diff is passed through a LangGraph of LLM-powered agents in sequence
• Skips agents that are not applicable (e.g., no UI or test files)

---

📦 Installation

1. Clone the repo:

   cd commit-intel

2. Set up a virtual environment:

   python -m venv .venv
   source .venv/bin/activate
   pip install -e .

3. Pull the model using Ollama:

   ollama pull codellama:7b

4. Run the CLI:

   review --help

---

💻 Example Output

review commit

🧠 Commit Summary:
- Updated login flow to use new auth provider
- Removed legacy authentication code

🧪 Critique:
- Missing error handling for network failure
- Redundant logging in auth callback

💡 Suggestions:
- Extract token parsing into a separate function

---

📁 Project Structure

commit-intel/
├── cli/                # Typer CLI entrypoints
├── core/               # LangGraph, Ollama, agents
├── .pre-commit/        # Hook logic
├── tests/              # Unit tests (TBD)
└── README.md

---

🔮 Future Improvements (Phase 3+)

• ✅ Export feedback to Markdown (--markdown)
• ✅ CLI flags to enable/disable agents
• ⚙️ Gradle plugin integration
• 📈 Coverage integration from JaCoCo reports
• 🔁 Streaming token-based output
• 🔧 Model config via .commit-intel.yml
• 🧠 Model fallback: use OpenAI if Ollama is down
• 🛠️ Workspace scoring: code quality % per commit
• 🧪 Automated fix suggestions (optional auto PRs)

---

🧠 Credits

Built by Android engineers, for Android engineers.
Inspired by LangChain, LangGraph, Ollama, and open-source LLM tooling.

MIT License.