What I Built
Touch Grass Bot — a small Python command-line tool that nudges you outside instead of letting you keep scrolling. You tell it how you’re feeling or how much free time you have (e.g. “tired, got 50 minutes”), and it suggests one simple, relevant outdoor activity. It’s built for anyone stuck in a screen-loop who just needs a quick, low-effort push to actually go outside.
Demo
Code
https://github.com/daralord/touch-grass-bot
How I Built It
I’m new to Python and to working with AI models directly — this was my first project actually running one. I used Qwen2.5-0.5B-Instruct, a small open-weight chat model from Alibaba, run fully locally through Hugging Face’s transformers library — no API key, no cloud calls after the initial model download. I built it step-by-step with Claude guiding me through the Python and explaining each part as I went, since this was new territory for me — from writing the script to debugging a few model compatibility issues along the way. The script takes the user’s input, wraps it in a simple instruction prompt (“suggest ONE outdoor activity based on this”), and runs it entirely on my own laptop using Python.
Why Does Open Innovation Matter?
Running this locally meant no API costs, no rate limits, and no dependency on an internet connection once the model was downloaded — something a closed, paid API like GPT-4 can’t offer. It also let me actually see and understand how the model processes input and generates a response, rather than treating it as a black box behind someone else’s server. As a student experimenting for the first time, that transparency and zero cost made the difference between being free to try things out and having to ration every API call.
Top comments (1)
You are doing great 👍