I'm new to local AI agents and want to experiment with one that can inspect an entire Visual Studio solution and explain how each project works in user-friendly language. The solution mainly contains VB.NET client applications, with an MSSQL project included. Ideally, the agent would read the projects and produce Markdown documentation describing workflows and controls—for example, explaining that pressing a green button displays a message box for the selected object.
I tried Codex Sol medium and was impressed with the results, but the real codebase is confidential and can't be sent to a cloud service. I have an M3 Max available for running local models. What would be a practical way for a beginner to set this up?
2 Answers
Qwen 2.5 Coder 32B is a reasonable local option for this kind of code-reading task. You can run it through Ollama and connect it to an editor agent such as Continue in VS Code. Point the agent at each project folder, then use a system prompt that asks it to focus on user-visible behavior—especially event handlers, button actions, forms, validation, and database interactions—and write the results as Markdown. For a large solution, process the projects and files in smaller batches rather than asking the model to understand everything at once.
A local Qwen model should be enough to get started. Ollama is one of the simplest ways to host the model locally because it provides an API that tools like Continue can use. Start by testing it on a single project, asking it to identify forms, controls, event handlers, and important data flows before generating the final user documentation. Once the output is reliable, automate the process project by project and have the agent save each explanation to a separate Markdown file.
Do I need to run the model through Ollama, or is there another setup I should use? I’m completely new to running these models locally.