clean-build-studio/local-ai-work-agent-template — reverse-engineered prompt

Build me a small Python starter project for a local AI work agent that runs completely on my computer using Ollama. I want it to read settings from a local env file, use Qwen3 4B by default, connect to Ollama at localhost, and print a simple test response so I know the model is working. Please include safe defaults, with full computer access off, file access on only for a project data folder, and output saving enabled.

Set up folders for an inbox, processed files, and outputs, with the app ready to grow later with connectors and processors. Keep the first version simple, mostly just proving Python can talk to the local model and return a clean answer without extra thinking text. Add an example env file, requirements, git ignore rules, and a clear README that walks a beginner through installing Ollama, pulling the model, creating a virtual environment, installing packages, and running the first test.