code-lm 0.3.0

An AI coding assistant using various LLM models.

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

Model	ID	Context	Strength
NVIDIA Nemotron 3 Super 120B	nvidia/nemotron-3-super-120b-a12b:free	262K	Default
Qwen3 Coder 480B	qwen/qwen3-coder:free	262K	Best coding
OpenAI GPT-OSS 120B	openai/gpt-oss-120b:free	131K	Open source
Llama 3.3 70B	meta-llama/llama-3.3-70b-instruct:free	66K	General purpose
Mistral Small 3.1 24B	mistralai/mistral-small-3.1-24b-instruct:free	128K	Vision + tools
Qwen3 Next 80B A3B	qwen/qwen3-next-80b-a3b-instruct:free	262K	Tool calling
NVIDIA Nemotron Nano 9B v2	nvidia/nemotron-nano-9b-v2:free	128K	Fast
Trinity Large Preview	arcee-ai/trinity-large-preview:free	131K	Reasoning
Trinity Mini	arcee-ai/trinity-mini:free	131K	Fast & efficient
GLM 4.5 Air	z-ai/glm-4.5-air:free	131K	Tool calling
Step 3.5 Flash	stepfun/step-3.5-flash:free	256K	Tool calling
MiniMax M2.5	minimax/minimax-m2.5:free	197K	Tool calling
OpenAI GPT-OSS 20B	openai/gpt-oss-20b:free	131K	Lightweight
Gemma 3 27B	google/gemma-3-27b-it:free	131K	Vision
Qwen3 4B	qwen/qwen3-4b:free	41K	Ultra lightweight
Hermes 3 Llama 3.1 405B	nousresearch/hermes-3-llama-3.1-405b:free	131K	Large Llama
Llama 3.2 3B	meta-llama/llama-3.2-3b-instruct:free	131K	Tiny & fast

---

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