This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass
What I Built
TouchGrass is an AI-powered outdoor quest generator with one unusual goal: to make you stop using it.
You tell it four things:
- How much time you have
- Your current mood
- How much you want to spend
- How adventurous you feel
It generates a small, personalized real-world quest. For example:
The Stranger
30 minutes · Free · EasyExplore somewhere nearby you've never noticed before. Find something you've never noticed, take a photo, and spend a few quiet minutes observing your surroundings.
Then you close the app and actually go do it.
Most AI products are designed to keep you engaged: more messages, more prompts, more time inside the app. TouchGrass flips that. The ideal user journey is:
Open the app → Tell it what you're up for → Get a quest → Go outside → Touch grass.
If you spend less time on the website, it has done its job.
It's for anyone who spends too much time on a screen and wants a nudge (and a concrete idea) to get outside, without needing to plan anything or spend much money.
Demo
🌱 Live demo: touchgrass.adhithyan.org
Code
Adhithyan2004
/
touchgrass
TouchGrass is an AI-powered outdoor quest generator that creates personalized real-world activities based on your time, mood, budget, and preferred difficulty.
🌱 TouchGrass
An AI whose primary goal is to make you stop using the app.
TouchGrass is an AI-powered outdoor quest generator that creates personalized real-world activities based on your time, mood, budget, and preferred difficulty.
The idea is simple:
Open the app → Generate a quest → Close the app → Go outside. 🌱
🔗 Live Demo: https://touchgrass.adhithyan.org
🏆 Hacktoberfest 2026 — Week 1
TouchGrass was built as my submission for the Hacktoberfest Week 1 "Touch Grass" challenge.
The challenge is about building something with open-source AI at its core while encouraging people to spend more time in the real world.
TouchGrass takes that idea literally.
The website's job is to get you off the website.
✨ What It Does
Tell TouchGrass:
- ⏱️ How much time you have
- 🌈 Your current mood
- 💰 How much you want to spend
- 🧗 How adventurous you feel
The AI then generates…
How I Built It
The whole point of this challenge is open-source AI at the core, so I wanted the AI to run on infrastructure I control rather than calling a proprietary cloud API.
The AI layer
- Ollama for local model inference
- Gemma, an open-weight model, generates the quests
The app
- Next.js, React, TypeScript, and Tailwind CSS
- A single API route (
/api/quest) takes the user's preferences, prompts the model, and returns a quest
Deployment
TouchGrass runs on my own Ubuntu home server with Docker, with Ollama running alongside the app. It's exposed to the internet through a Cloudflare Tunnel, with Nginx in front.
User
│
▼
touchgrass.adhithyan.org
│
▼
Cloudflare Tunnel
│
▼
Ubuntu Home Server
│
├── TouchGrass
│ │
│ ▼
│ Ollama
│ │
│ ▼
│ Gemma
│
└── Other self-hosted services
Keeping it safe
I didn't want to blindly trust a small model's output, so quests are constrained and checked. They're designed to be achievable by ordinary people, to respect the requested time and budget, and to avoid dangerous activities, trespassing, special equipment, and anything that needs internet access. The model is told to use generic nearby public locations instead of inventing specific places. The app also validates the AI's output and adds deterministic safety information on top.
Running it yourself is simple: install Ollama, pull a Gemma model, set OLLAMA_URL in .env.local, and run npm run dev.
Why Does Open Innovation Matter?
TouchGrass wouldn't be the same project if the AI were simply a call to a closed API.
Using an open-weight model allowed me to run the AI myself on hardware I control. The entire quest-generation process can happen without sending user preferences to a proprietary AI provider.
More importantly, open models make experimentation accessible. I could take a small model like Gemma, run it on an old laptop sitting at home, connect it to a real application, and actually deploy the result.
That's what makes the project interesting to me: I didn't need a huge cloud infrastructure to build something with AI at its core. I could run the model myself and build around it.
And in a project whose entire purpose is to get people away from their screens, having the AI itself run locally felt especially fitting.
Prize Categories
- Best Use of Gemma — TouchGrass uses Google's Gemma open-weight model through Ollama to generate personalized outdoor quests based on the user's time, mood, budget, and difficulty preferences.
Top comments (0)