TouchGrass AI: An Open-Source AI Companion That Gets You Off the Screen and Into the Real World

Abhinav Kumar — Wed, 07 Oct 2026 19:34:53 +0000

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

🌿 TouchGrass AI — Less Screen. More World.

What I Built

I built TouchGrass AI, a web-based outdoor companion designed around a simple idea:

What if AI encouraged us to use our screens less instead of keeping us on them?

TouchGrass AI creates personalized outdoor adventures based on the user's:

📍 Location
🎯 Activity
⏱️ Available time
💪 Difficulty

The user can choose activities such as:

🌿 Nature Walk
🥾 Hiking
🏃 Running
🐦 Bird Watching

The AI then creates a set of outdoor missions.

For example, a Nature Walk might include:

Walk for 10 minutes without checking your phone.
Find three different leaf shapes.
Stop somewhere quiet and listen for 60 seconds.
Find something you've never noticed before.
Take a moment to observe your surroundings.

The project also includes Phone Down Mode, where the interface becomes a minimal full-screen experience with a timer and one clear instruction:

📵 PHONE DOWN — Go outside. Look around. Be present.

After the adventure, TouchGrass AI generates an Adventure Report with a Grass Score, completed missions, discoveries, and an AI-style reflection.

The target users are students, runners, hikers, nature lovers, and anyone who wants a reason to step away from their screen and spend more time outdoors.

Demo

🌐 Live Website:
https://abhinavkumar-3676.github.io/TouchGrass-AI/

The website is deployed using GitHub Pages and can be accessed directly from a browser without installing anything.

Code

💻 GitHub Repository:
https://github.com/Abhinavkumar-3676/TouchGrass-AI

The project is built using a lightweight frontend stack:

HTML
CSS
JavaScript

There is no framework or complicated build system required.

The goal was to keep the project simple, accessible, and easy for other developers to understand and contribute to.

How I Built It

TouchGrass AI is built as a frontend-first web application using HTML, CSS, and JavaScript.

The main components are:

🌿 AI Adventure Generator

The application takes the user's activity, duration, difficulty, and location and generates a personalized outdoor adventure.

📵 Phone Down Mode

A full-screen mode designed specifically to minimize interaction with the website once the adventure begins.

The idea is intentionally unusual:

A successful session should result in the user spending less time on the website.

⏱️ Adventure Timer

Users can start, pause, and reset their outdoor activity timer.

🌱 Nature AI

Users can describe something they saw or discovered outside and receive AI-assisted guidance about what to observe.

🏆 Grass Score

The application calculates an adventure score based on completed missions and turns the experience into a simple progress system.

🧠 Open AI Direction

The project is structured so that its AI functionality can be connected to an open-weight AI model rather than being tightly coupled to one closed AI provider.

Possible open-weight models for the project include:

Qwen
Llama
Gemma

This makes it possible to experiment with different models and eventually move toward local inference.

Why Does Open Innovation Matter?

Open innovation matters because an outdoor AI application has some requirements that are different from a normal chatbot.

🔐 Privacy

An outdoor companion can potentially deal with sensitive information such as:

Location
Photos
Activity history
Personal observations

Using open models and local inference can make it possible to process this information without depending entirely on a closed AI service.

🌐 Offline Potential

One of the biggest future goals for TouchGrass AI is offline AI.

Imagine being on a hiking trail with:

No Wi-Fi
Weak mobile signal
No cloud API access

A lightweight open-weight model running locally could still generate missions or provide basic assistance.

That makes open AI especially interesting for outdoor applications.

🔄 Model Freedom

With open-weight models, developers can experiment with different models instead of being locked into a single provider.

The same application could potentially use:

Qwen
   ↓
Llama
   ↓
Gemma

depending on the device, performance requirements, and use case.

💰 Accessibility

Open models can also reduce dependency on per-request closed API costs, especially when models can be run locally.

For a project intended to encourage people to explore freely, having an AI architecture that can eventually work without a constant cloud connection is a major advantage.

My Agent Session

I did not include an agent session in this submission.

Prize Categories

🧠 Open-Source AI

TouchGrass AI is designed around the use of open-weight AI and the ability to experiment with different open models.

🌿 Touch Grass

This is the primary category for the project.

The entire concept is based on using AI to encourage people to leave the screen and interact with the real world.

The Idea Behind TouchGrass AI

Most digital products are optimized for:

More engagement. More screen time.

TouchGrass AI tries to optimize for the opposite:

Less screen time. More real-world experience.

The ideal user journey is:

Open TouchGrass AI
        ↓
AI creates an adventure
        ↓
Phone Down Mode
        ↓
Go outside
        ↓
Explore
        ↓
Complete missions
        ↓
Return
        ↓
Get your Adventure Report

The best outcome isn't spending more time inside the app.

It's closing the app and going outside. 🌿

🔗 Links

🌐 Live Demo:
https://abhinavkumar-3676.github.io/TouchGrass-AI/

💻 GitHub:
https://github.com/Abhinavkumar-3676/TouchGrass-AI

Final Thought

AI doesn't always have to give us more things to look at. Sometimes, its job should be to tell us to put the screen down and look at the world. 🌎🌿

DEV Community: Abhinav Kumar