Player not loading? Watch on YouTube
This tutorial builds a local code auditor with QwenPaw and an Ollama backend running Qwen 2.5 3B. The project checks a sample Python file and produces a Markdown report plus an HTML dashboard. It is a practical example of using a local LLM to turn code metrics and security findings into recommendations.
The setup covers a pip installation, Python dependencies from requirements.txt, model setup in Ollama, and execution with python main.py. The speaker explains how main.py assembles an audit prompt, sends it to the local Ollama API, and writes the results into report templates.
In the demonstrated run, the sample has 44 lines of code, three functions, maximum complexity of eight, and a reported security score of 50 out of 100. Two findings concern os.system and execution with shell=True. The generated recommendations call for removing those shell execution risks and refactoring deeply nested control flow. The browser report displays severity colors and includes a Markdown download button.
The opening overview describes QwenPaw as an AI agent with layered memory, parallel agents, and a coding interface. The speaker also claims persistent conversation recall and kernel-level sandbox protection. The code audit demonstrates report generation; it does not establish those broader memory or security guarantees.
