What I Built
I built TouchGrass AI, a small open-source AI-powered outdoor activity planner designed to help people spend less time on screens and more time outside.
The idea is simple: instead of spending a long time deciding what to do, the user gives a few details such as their available time, preferred activity, and energy level. The application then generates a simple outdoor mission that they can complete in the real world.
The goal is to make the interaction with the application short and actionable — use the screen to get the mission, then put the phone away and go outside. 🌱
Why I Built It
It is very easy to spend too much time on screens, even when we actually want to take a break.
I wanted to experiment with whether AI could be used for the opposite purpose: instead of keeping someone on a screen, use AI to give them a reason to leave the screen.
For example, the application can suggest activities such as:
- Going for a short walk
- Exploring a nearby area
- Sitting outside and observing nature
- Taking a short outdoor break
- Going on a photo walk
- Finding something interesting around you
The generated mission is intended to be simple enough that the user can immediately act on it.
Demo
Try TouchGrass AI here:
https://harshitha-hg.github.io/Touchgrass-ai/
Code
The complete source code is available on GitHub:
https://github.com/Harshitha-HG/touchgrass-ai
How I Built It
I built the project using:
- HTML
- CSS
- JavaScript
- WebLLM
- SmolLM2 open-weight model
- GitHub Pages
The application uses WebLLM to run an open-weight AI model directly in the browser.
The model is used to generate personalized outdoor missions based on the user's inputs.
The project also includes a fallback mechanism so that the application can still provide a useful outdoor mission on devices where local AI inference has limitations.
How Open-Source AI Is Used
The AI is not just added as a chatbot.
The open-weight model is part of the core functionality of the project because it generates the outdoor mission.
The project uses an open-weight SmolLM2 model through WebLLM, allowing inference to happen in the browser rather than requiring a traditional cloud AI API for the main generation process.
This makes the project interesting from an open-AI perspective because developers can inspect the implementation, modify the prompts, experiment with different open-weight models, and extend the application.
How It Gets People Outside
The main idea behind TouchGrass AI is:
Short interaction → Get a mission → Put the phone away → Go outside.
Instead of creating another application that encourages users to continuously interact with their screens, TouchGrass AI tries to make the screen interaction as short as possible.
The output is an activity that can be completed away from the device.
Why Open Innovation Matters
Open innovation makes it possible for developers to understand, modify, and improve the technology they use.
With open-source AI and open-weight models, developers can experiment with different models, change prompts, build new applications, and learn from the underlying technology.
For a project like TouchGrass AI, this also makes experimentation easier. Someone could fork the project and turn it into:
- A bird-watching assistant
- A running route planner
- A gardening assistant
- A nature exploration tool
- A local photography challenge generator
The community can take the basic idea and build something completely different from it.
What I Learned
This project helped me understand how open-weight AI models can be integrated directly into a web application.
I also learned that an AI project does not have to be complicated to be useful.
The interesting part of TouchGrass AI is not simply generating text with AI. The purpose is to use AI to encourage an activity that happens outside the screen.
Final Thought
AI is usually associated with keeping us connected to our devices.
With TouchGrass AI, I wanted to try something different:
Use AI to help you put the phone down. 🌿
Top comments (0)