Local LLM
code-context-engine
Code-context-engine, in Python — built for save 94% on ai coding tokens. index your cod...
★ 425 starsMIT
View guide →Qwen3-Coder - Alibaba's open code model series for agentic coding, up to 1M tokens of context
Qwen3-Coder is an open-source local llm project by QwenLM (QwenLM/Qwen3-Coder on GitHub). It has 16,838 GitHub stars and 1,256 forks, is released under the NOASSERTION license, and was last updated on GitHub on 24 March 2026.
Choose a Qwen3-Coder size that fits your GPU, or use it through an agent tool or hosted API.
Clone this repo for docs and examples, and download the weights from the linked model pages.
Load tokenizer and model with the README's quick-start pattern and apply the chat template.
Generate a completion, try fill-in-the-middle, or point an agent platform at the model.
Designed for agentic coding, agentic browser use and foundational coding tasks, with results the README compares to Claude Sonnet among open models.
Supports platforms such as Qwen Code, Cline and Claude Code, with a specially designed function-call format.
Native 256K token context, extendable up to 1M tokens with YaRN - sized for repository-scale understanding.
Supports FIM completion for IDE-style code infilling, shown in the quick start.
Uses the apply_chat_template flow with ChatML, consistent with earlier Qwen releases.
The README demonstrates full tasks - releasing a website, building a parkour game, sound ASCII art and more.
Get the source code from GitHub and enter the project directory.
git clone https://github.com/QwenLM/Qwen3-Coder.git
cd Qwen3-CoderThe README quick start loads the tokenizer and model, formats messages with the chat template and generates.
# Follow the Quick Start code in the README (load model, apply_chat_template, generate)Configure Qwen3-Coder as the model in Qwen Code, Cline or Claude Code per the README's platform notes.
# Point your agent platform at the Qwen3-Coder endpoint or local server - see READMEIt is maintained by Alibaba's Qwen team (QwenLM) on GitHub.
The repository content is freely available; the model weights carry the license stated on their model cards - check there before commercial use, as the README text itself does not state one.
No API key is needed to run the weights locally. Hosted endpoints and agent platforms you use it through have their own accounts and pricing.
Compared with general Qwen3 models, Qwen3-Coder is specialized for code and agentic tool use; compared with proprietary coding models, it is open and self-hostable with very long context.
Clone the repo, read the Quick Start, download a checkpoint, and run the chat-template example - or select it inside Qwen Code or Cline.
Code-context-engine, in Python — built for save 94% on ai coding tokens. index your cod...
DeepSeek-Coder-V2 — DeepSeek's open code model family that challenged closed-source code models.
StepFun's open 196B MoE model that activates 11B parameters per token, built for fast agentic reasoning.
Local code-intelligence MCP server and CLI for coding agents.
Local codebase intelligence CLI and MCP for coding agents.
Local-first dashboard of tokens, models, and cost for coding agents.
The most-starred other Local LLM projects in the LibreAI catalogue — the realistic shortlist if Qwen3-Coder is not the right fit.
The simplest way to run open language models on your own machine.
User-friendly AI Interface (Supports Ollama, OpenAI API, ...)
LLM inference in C/C++
A high-throughput and memory-efficient inference and serving engine for LLMs
🤯 LobeHub is your Chief Agent Operator, organizing your agents into 7×24 operations by hiring, scheduling, and reporting on your entire AI team.
Free and open — grab the source, star the repo, and start creating.