git-auto-shelldm 🤖💻

An intelligent, context-aware command-line assistant that translates natural language into Git commands. Powered by Qwen 2.5 Coder 32B and the Hugging Face Inference API, it provides a lightning-fast, secure, and interactive shell designed to optimize your Git workflows.

✨ Key Features

• 🚀 Near-Instant Startup: Optimized to boot in under 2 seconds (3.7x faster than traditional vector-db CLI tools) by using a lightweight, pure-Python memory system instead of heavy local model engines.
• 📦 Global Installation: Install with a single command via NPM and run it anywhere in your terminal using the git-autopilot command.
• 🔑 Secure Token Onboarding: Prompts for your Hugging Face API token on first startup and caches it safely in your home directory (~/.git_autopilot_config.json) so you don't have to keep pasting it.
• 🛡️ Human-in-the-Loop Safety Gate: Prompts you for confirmation before executing any generated Git command, protecting your repository from unintended changes.
• 🧠 Global Semantic RAG Memory: Remembers past successful commands and searches your local cheat sheet. Learning is stored globally in your home directory (~/.git_autopilot_memory.json) and shared across all of your local git repositories.
• 🔄 Auto-Correction Loop: If a command fails to execute, the assistant automatically captures the error output, analyzes it, and attempts to correct and execute it (up to 3 retries).
• 📦 Git LFS Support: Fully integrated with Git Large File Storage (LFS) commands.

🚀 Quick Start

1. Install Globally

Install the CLI command globally on your system using NPM:

npm install -g git-auto-shelldm

(Requires Node.js and Python 3 to be installed on your system)

2. Run the Assistant

Launch the interactive shell from within any Git repository:

git-autopilot

3. Enter your HF_TOKEN (First run only)

If you do not have the HF_TOKEN environment variable set, the assistant will guide you to save it:

Welcome to Git Auto-Pilot!
To use the Hugging Face API, you need a free API token.
Get your token here: https://huggingface.co/settings/tokens
Enter your Hugging Face API Token (HF_TOKEN):

💡 Usage Examples

Type your request in plain English:

Git-Bot> initialize a new repository
Thinking...
Proposed Command: git init
Execute this command? [y/N]: y

Git-Bot> track all large ZIP files in the repository
Thinking...
Proposed Command: git lfs track "*.zip"
Execute this command? [y/N]: y

🛠️ How It Works (Architecture)

1. Lightweight Embedding Retrieval: Queries the user input embedding using the Hugging Face feature extraction model (sentence-transformers/all-MiniLM-L6-v2) and matches it against your global memory database (~/.git_autopilot_memory.json) using a pure-Python cosine similarity function.
2. Context Injection: Relevant matching commands from the cheat sheet or history are injected into the model's RAG context.
3. Qwen 2.5 Coder LLM: The user's query, system path, OS type, and RAG context are sent to Qwen/Qwen2.5-Coder-32B-Instruct to output the exact Git command.
4. Command Execution & Cache: Upon confirmation, the command is executed. If successful, the input and output are cached locally in the database.