LM Code

LM Code is a powerful AI coding assistant for your terminal supporting 17 free models via OpenRouter. Chat with AI models, automate file operations, and boost your workflow directly from the command line.

Features

• Interactive CLI with AI Assistance:
  • Chat with AI models for coding advice, file management, and more.
  • Markdown rendering for improved readability.
• 17 Free Models via OpenRouter:
  • NVIDIA Nemotron, Qwen, OpenAI, Meta Llama, Mistral, and more.
• Automated Tool Usage:
  • File operations: view, edit, grep, glob.
  • Directory operations: ls, tree, create_directory.
  • System commands: bash.
  • Quality checks: linting, formatting.
  • Test running: pytest and similar tools.
• Customizable Configurations:
  • Easily set default models and API keys.

Installation

Method 1: Install from PyPI (Recommended)

pip install code-lm

Method 2: Install from Source

git clone https://github.com/Panagiotis897/lm-code.git
cd lm-code
pip install -e .

Setup

Before using LM Code, set up your API key for OpenRouter.

lmcode setup YOUR_OPENROUTER_API_KEY

This saves your API key in ~/.config/lm-code/config.yaml.

Usage

Start an Interactive Session

# Start with the default model (NVIDIA Nemotron 3 Super 120B)
lmcode

# Start with a specific model
lmcode --model qwen/qwen3-coder:free

Manage Models

# Set a default model
lmcode set-default-model qwen/qwen3-coder:free

# List all available models
lmcode list-models

Supported Models

Interactive Commands

During an interactive session:

• /exit: Exit the session.
• /help: Display help information.

How It Works

LM Code uses native tools to enhance your coding experience:

1. You ask: "What files are in the current directory?"
2. LM Code uses the ls tool to fetch directory contents.
3. The assistant formats and presents the response.

Development

LM Code is under active development. Contributions, feature requests, and feedback are welcome!

Changelog

v0.3.0

• Updated default model to NVIDIA Nemotron 3 Super 120B.
• Added 17 free models from OpenRouter (previously 6).
• Fixed ModuleNotFoundError: No module named 'gemini_cli' from stale entry point.
• Fixed UnicodeDecodeError on Windows (cp1253) for all subprocess commands.
• Fixed API URL (/chat/completions was missing) causing HTML response errors.
• Improved API error handling for empty/invalid responses.
• Added API key validation on startup.
• Tool parameter descriptions now use args_schema when available.
• Removed conflicting logging.basicConfig from openrouter module.
• Added UTF-8 encoding with fallback to all subprocess calls.

v0.2.5

• Added more models to the model list.
• Fixed crucial bugs from previous versions.
• Removed Gemini models.
• Updated models to latest versions.

v0.1.0

• Rebranded from Gemini to LM Code.
• Integrated OpenRouter as the default provider.
• Added multi-model support.
• Overhauled CLI commands (gemini -> lmcode).

Future Plans

• Pricing with appropriate rate limits.
• Non-free model support.
• MCP Server integration.
• Additional providers.

Known Issues

• If you used earlier versions, you may need to delete your old configuration:

  rm -rf ~/.config/lm-code
  

License

MIT License