🌱 I Built an AI That Tells You to Stop Using AI

This is a submission for the Hacktoberfest Open-Source AI Challenge: Week 1 — Touch Grass

🌱 TouchGrass AI: An AI That Tells You to Stop Using AI

What if the best thing an AI could do was tell you to put your screen away?

That's the idea behind TouchGrass AI, a lightweight local AI-powered outdoor companion I built for Hacktoberfest 2026 Week 1.

The goal is simple:

Use AI for a few seconds. Get a mission. Put the screen away. Go outside.

No endless chatbot conversation. No account required. No cloud database.

Just a small push to spend some time in the real world.

What I Built

TouchGrass AI generates short outdoor missions based on the activity you choose and the amount of time you have.

You can choose activities such as walking, exploring, gardening, birdwatching, or nature photography.

🏠 The Home Screen

The user chooses an activity and how much time they have.

🌳 The AI-Generated Mission

For example, a 10-minute walking mission can give the user:

10-Minute Walking Mission

Go outside and spend 10 minutes exploring.

Notice three things you normally overlook.

Spend one quiet minute listening to your surroundings.

The important part is that the AI interaction is intentionally short.

The screen should be the shortest part of the experience.

⏱️ The Outdoor Session

The user can start the mission and put the screen away while spending time outside.

💭 Reflection

After completing the mission, the user can record what they noticed or experienced.

📖 Local History

Previous missions and reflections can be viewed locally.

I also tested the app outside myself, and I found it useful as a small push to spend time with nature.

The AI doesn't need to produce something complicated.

A simple activity and a specific amount of time can be enough to turn:

"I should go outside."

into:

"Okay, I'll spend the next 10 minutes outside."

Demo

🎥 Watch the full demo video:

https://drive.google.com/file/d/1A2MAGD9iCRS_q-vSq14IhM0CTXtGfRtZ/view?usp=sharing

The demo shows the complete flow:

Choose activity
↓
Choose available time
↓
Generate outdoor mission
↓
Start outdoor session
↓
Go outside 🌳
↓
Return
↓
Write reflection
↓
View local history

Code

The complete project is open source on GitHub:

https://github.com/codelift338-dev/touchgrass-ai

The repository contains:

Flask backend

Frontend

Requirements

README

Git configuration

The GGUF model itself isn't committed because the model file is hundreds of megabytes. It is downloaded separately and loaded locally by llama.cpp.

How I Built It

The most important design decision was making open-source/local AI the core of the application, rather than simply connecting a closed AI API.

The main technology stack is:

Qwen 2.5 0.5B Instruct — open-weight language model

GGUF quantization — lightweight model format

llama.cpp — local inference

Python

Flask

HTML/CSS/JavaScript

Browser localStorage — local reflections and history

The architecture looks like this:

User
│
▼
TouchGrass AI Web App
│
▼
Flask Backend
│
▼
llama.cpp
│
▼
Qwen 2.5 0.5B
│
▼
Outdoor Mission
│
▼
Go Outside 🌳
│
▼
Local Reflection

I built and tested the project on a fairly modest Windows computer:

8 GB RAM

Intel Pentium 2.16 GHz

Intel HD Graphics

Because the computer doesn't have a dedicated GPU, I chose a very small quantized model instead of trying to run a much larger local model.

That turned out to be a good fit.

TouchGrass AI doesn't need to write an essay or solve a complicated problem.

It just needs to give someone a useful reason to step outside.

Why Does Open Innovation Matter?

This is where local open-source AI made a real difference.

I could have built TouchGrass AI around a hosted AI API, but I wanted the application to be more independent and privacy-friendly.

🔒 Privacy

Outdoor reflections can be personal.

TouchGrass AI doesn't require an account or cloud database.

The reflection and mission history are stored locally in the browser.

💻 Local Control

The AI runs on the user's own computer through llama.cpp.

The application isn't dependent on a proprietary AI service for every generation.

💰 No Per-Request API Cost

Once the local model is downloaded, inference happens on the user's machine.

There isn't a cloud AI API charge for every outdoor mission.

🔄 Model Choice

Because the application uses a local GGUF model, the model can be replaced or upgraded without rebuilding the entire application around a closed provider.

🪶 Lightweight AI

Open-weight models made it possible for me to experiment with local AI even on modest hardware.

That was particularly important because I built this project on an 8 GB RAM machine with an Intel Pentium processor and integrated graphics.

The Part I Like Most

There is a funny contradiction at the heart of this project.

I'm using AI to tell people to stop using their screens.

But that's exactly the point.

I don't want TouchGrass AI to become another application that demands attention.

The ideal interaction is:

Open app
↓
Get mission
↓
Put screen away
↓
Go outside
↓
Experience something
↓
Come back
↓
Reflect

The AI should disappear as quickly as possible.

The model provides the suggestion. Nature provides the experience.

What's Next?

There are several things I could add:

Weather-aware outdoor missions

Seasonal activities

Local nature suggestions

More outdoor activity categories

Difficulty levels

Printable outdoor mission cards

Better offline-first support

Support for additional lightweight local models

But I want to keep the core idea simple.

The goal isn't to build another platform that keeps people staring at a screen.

The goal is to make the screen useful for a moment — and then help someone leave it behind.

🏆 Hacktoberfest 2026 — Week 1: Touch Grass

This project was built for Hacktoberfest 2026 Week 1: Touch Grass.

The challenge asks participants to build something with open-source AI at its core that encourages people to get outside.

TouchGrass AI is my attempt to do exactly that:

Build an AI that helps you spend less time with AI. 🌱