TouchGrass AI: Turn AI Prompts into Real-World Adventures 🌿

A developer built TouchGrass AI, a local AI-powered outdoor mission generator that converts a user's mood, available time, preferred activity, and difficulty into real-world activities such as walking challenges, photography tasks, and birding. The project runs Qwen 3 4B locally through Ollama behind a React and Tailwind frontend with a Node.js and Express backend, returning structured JSON missions that include a goal, challenges, a phone-down rule, duration, difficulty, and safety guidance. The developer says the open-weight, local-inference approach avoids sending personal prompts to a third-party provider and sidesteps API costs and usage limits.

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass https://dev.to/challenges/hacktoberfest-week1-2026-10-05 TouchGrass AI is a local AI-powered outdoor mission generator that turns a user’s mood, available time, preferred activity, and difficulty into a simple real-world mission. Instead of giving users another screen-based AI experience, it uses Qwen 3 4B through Ollama to create practical activities such as walking challenges, nature exploration, photography tasks, birding, mindfulness, and outdoor mini-adventures. The goal is simple: use AI to help people stop scrolling and start doing. Each generated mission includes a clear goal, challenges, a phone-down rule, duration, difficulty, and safety guidance. It is designed for students, remote workers, and anyone who wants a quick reason to step outside. The AI inference runs locally, keeping the core experience private and demonstrating how open-weight AI can be used for something that ultimately happens away from the screen. https://hacktoberfest-touchgrassai.vercel.app/ https://hacktoberfest-touchgrassai.vercel.app/ https://github.com/RajBhokare/Hacktoberfest-Touchgrass ai https://github.com/RajBhokare/Hacktoberfest-Touchgrass ai How I Built It TouchGrass AI is built around local, open-weight AI, with Qwen 3 4B running through Ollama. The AI inference happens locally on the user’s machine rather than through a closed cloud AI API. The architecture is: React + Tailwind CSS → Node.js + Express → Ollama → Qwen 3 4B → Structured Outdoor Mission → React UI I built the frontend with React and Tailwind CSS, while the backend uses Node.js and Express to communicate with the local Ollama API. Users provide their mood, available time, difficulty, and preferred activity. These inputs are converted into a structured prompt and sent to Qwen 3 4B. The model returns structured JSON containing the mission title, duration, difficulty, description, challenges, phone-down rule, and safety note. The frontend then turns that response into an interactive mission experience. The project demonstrates how open-weight models + local inference can power useful applications without depending on proprietary AI APIs. Open innovation matters because it gives developers more control, transparency, and freedom to experiment with AI. For TouchGrass AI, using the open-weight Qwen 3 4B model through Ollama made it possible to run the core AI locally instead of depending on a closed API. This means the project can work without sending users’ personal prompts to a third-party AI provider, while also avoiding API costs and usage limits. I could directly control how the model is prompted, how its output is structured, and how it fits into the application. More importantly, open AI makes experimentation accessible. A student can download a model, build around it, modify the application, and understand how the pieces work without needing access to an expensive proprietary platform. For TouchGrass AI, that freedom helped turn AI from something that keeps people online into a tool designed to get them offline and into the real world. I used DevRelay during development to experiment with an agent-assisted workflow and iterate on the project. The session helped me move from the initial concept to a working local-AI application using Qwen 3 4B + Ollama, while keeping the implementation focused on open-source AI. If you didn't use DevRelay at all, use this instead: My Agent Session I did not use DevRelay for this project. I built and iterated on TouchGrass AI using an AI-assisted development workflow, with Qwen 3 4B running locally through Ollama as the core open-weight AI component.