This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass
What I Built
I built TouchGrass AI, an open-source AI companion designed to get people off their screens and into the real world.
The idea is simple:
Use AI for a few seconds. Then put the phone away. 🌱
Instead of building another AI assistant that encourages people to spend more time chatting with a model, TouchGrass AI generates small, practical outdoor missions based on:
- 🌿 Activity type
- ⏱️ Available time
- ⚡ Energy level
- 📍 Outdoor setting
- 🎯 Personal interests and goals
The app currently has three modes:
- Nature — activities focused on observing and exploring nature
- Explore — small outdoor exploration challenges
- Garden — simple gardening and plant-care activities
For example, a user can select:
Explore → 30 minutes → Low energy → Park
and receive a short outdoor mission designed to be completed away from the screen.
The goal isn't to keep the user inside the app.
The goal is to make the app unnecessary as quickly as possible.
The ideal interaction is:
10 seconds on the phone → 30 minutes outside. 🌳
Code
GitHub Repository: [https://github.com/Varun-666/touchgrass.ai]
The project is completely open source and includes the application code, AI integration, prompts, fallback system, frontend, backend, and documentation.
How I Built It
TouchGrass AI is built around an open-weight AI model rather than a proprietary AI API.
The AI layer uses:
Qwen2.5-0.5B-Instruct
The model runs locally through the Hugging Face Transformers ecosystem.
The basic architecture is:
User
│
▼
TouchGrass Web UI
│
▼
FastAPI Backend
│
▼
Local Qwen2.5-0.5B-Instruct
│
▼
Outdoor Mission
│
▼
📱 Put Phone Away
│
▼
🌳 Go Outside
The user provides a few simple preferences such as their available time, energy level, environment, and interests.
The local model then generates a structured outdoor mission containing things like:
- Mission title
- Short description
- Steps
- What to bring
- Safety notes
- A suggested duration
- A reason for doing the activity
I also included a fallback mission system so the application can still demonstrate its core experience if the AI model isn't available.
The project uses:
- Python
- FastAPI
- Hugging Face Transformers
- Qwen2.5-0.5B-Instruct
- HTML/CSS/JavaScript
- PWA support
No paid AI API or API key is required.
Why Does Open Innovation Matter?
This project is specifically designed around the idea that AI shouldn't always require sending everything to a remote server.
🔒 Privacy
Outdoor activities can involve photos, locations, gardens, homes, or other information about someone's surroundings.
A local AI architecture provides a path toward keeping that information on the user's own device instead of automatically sending it to a third-party AI provider.
📡 Offline Potential
People don't necessarily use TouchGrass AI while sitting next to a perfect internet connection.
The whole point is to use it while going outside.
Using an open-weight model means the project can be developed toward offline/local inference rather than depending entirely on an external API.
🔧 Model Freedom
Because the project isn't tied to a proprietary API, the AI component can be modified or replaced.
Developers can experiment with:
- Different open-weight models
- Quantized models
- Fine-tuned models
- Smaller models for mobile devices
- Different prompting strategies
The application isn't locked into one company's AI infrastructure.
💰 Accessibility
There is no per-request AI API cost.
That makes experimenting with the project much more accessible for students, hobbyists, and other developers who want to build AI applications without maintaining a paid API subscription.
The Interesting Part
Most AI applications try to maximize the amount of time users spend interacting with them.
TouchGrass AI tries to do the opposite.
A successful interaction isn't:
"I spent 30 minutes talking to the AI."
It's:
"The AI gave me something to do, and then I put my phone away."
That's the behavior I wanted to build around for this challenge.
My Agent Session
[ADD DEVRELAY AGENT SESSION HERE IF AVAILABLE]
Prize Categories
- Open-Source AI
- Touch Grass
Built with open-source AI for Hacktoberfest 2026. 🌱
Less screen. More outside.
Top comments (0)