TouchGrass AI: An Open-Source AI Companion to Help You Touch Grass

#devchallenge #hf26challenge #opensource #ai

🌱 TouchGrass AI — a little less scrolling, a little more living

live demo: https://touchgrass-ai.vercel.app/

github repository: https://github.com/arkendukundu-dev/touchgrass-ai

#hf26challenge #hacktoberfest #opensource #ai #react #python

🌍 the problem

we spend so much time looking at screens that sometimes we forget to experience the world outside them.

many of us want to spend more time outdoors, explore new places, build healthier habits, or simply take a break from endless scrolling. but knowing we should take a break isn't always enough.

i wanted to turn that simple idea into something fun and actionable.

🌿 introducing touchgrass ai

touchgrass ai is an open-source, ai-powered wellness app that encourages people to step away from their screens and reconnect with the real world through small, achievable activities.

instead of just telling you to spend less time online, it gives you little outdoor quests to complete in real life.

✨ what can you do with it?

🌱 daily quests: discover activities that encourage you to get outside and explore.
🤖 ai-powered suggestions: generate personalized activity ideas using open-source ai.
⭐ xp and levels: earn experience points as you complete quests.
🔥 streak tracking: build consistency through small daily actions.
📝 field notes: record thoughts and experiences from your offline adventures.
🎯 custom quests: create your own challenges.
📊 progress tracking: see your journey toward healthier digital habits.
📴 offline-friendly fallback: keep using built-in quests when ai is unavailable.

🛠️ how i built it

i used the following technologies:

frontend: react, vite, and css
backend: python and fastapi
open-source ai: qwen 2.5 3b through ollama
storage: browser local storage
deployment: vercel for the frontend and render for the backend

🧠 why open-source ai matters here

for a project focused on digital well-being, i didn't want the solution to depend entirely on a closed-source ai service.

using an open-weight model such as qwen 2.5 3b through ollama gives developers the freedom to experiment with local inference, customize model behavior, and explore different models without depending on a paid proprietary ai api.

local inference can also help keep prompts on a user's own machine when configured that way. the model is replaceable, which makes the project easier to experiment with and extend.

i also wanted the basic experience to remain useful without ai. the built-in quests provide a fallback, so the app's core idea doesn't disappear when a model is unavailable.

🚀 try it yourself

🌐 live demo: https://touchgrass-ai.vercel.app/

💻 open-source code: https://github.com/arkendukundu-dev/touchgrass-ai

the project is built with an extensible architecture, and there's plenty of room to improve quest personalization, add achievements, and introduce more ways to encourage offline exploration.