What If We Used AI to Spend Less Time on Screens?

TouchGrass AI — Get out. AI stays in.

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

What I Built

I built TouchGrass AI, a local-first AI outdoor companion that encourages people to spend less time on their screens and more time exploring the real world.

The idea behind the project is simple:

AI creates the experience, but the experience exists to get you away from AI.

Users can choose an outdoor activity, such as running, walking, hiking, or cycling, select a duration, and describe their goal. The AI then generates a personalized outdoor mission with a warm-up, main activity, outdoor challenge, cool-down, and motivation.

But generating a mission is only the beginning.

TouchGrass AI includes a Go Outside Mode with a timer so users can put their phones away and focus on their activity. When they return, they can reflect on their energy, mood, and difficulty, then save their experience to My Journey.

The app keeps experience history in the browser using local storage.

I designed it for students, developers, remote workers, and anyone who wants a little encouragement to step outside.

My goal wasn't to build another AI chatbot that keeps people engaged with a screen. I wanted to build something that gives people a reason to leave it.

Demo

Watch the project demo:

Touchgrass_AI.mp4 - Google Drive

drive.google.com

Code

GitHub repository: https://github.com/parthverma2005/Touchgrass_AI.git

TouchGrass AI is open source under the MIT License. The repository includes the frontend, FastAPI backend, setup instructions, and local AI integration.

How I Built It

I built TouchGrass AI using a lightweight web stack and an open-weight language model running locally.

Tech stack

• AI model: Qwen 2.5 0.5B
• Local inference: Ollama
• Backend: Python and FastAPI
• Frontend: HTML, CSS, and JavaScript
• Experience history: Browser localStorage

How it works

1. The user chooses an activity, duration, and goal.
2. The frontend sends the request to the FastAPI backend.
3. The backend sends a prompt to Qwen 2.5 0.5B through Ollama.
4. The generated mission is displayed in the frontend.
5. The user starts Go Outside Mode and completes the activity.
6. Afterward, the user records a reflection, which is saved locally in the browser.

I chose a small model because I wanted to experiment with local AI without requiring a powerful machine or a paid AI API.

The application does not require a cloud AI API for its core mission-generation feature. The model must be installed locally first, and the backend must be running for mission generation to work.

I also added a Local AI information panel to make the model and inference approach visible to users.

Why Does Open Innovation Matter?

Open innovation made this project possible in a way that fits its purpose.

Privacy by design

Mission generation can happen locally instead of sending each request to a cloud AI provider. The user's saved reflections also remain in their browser rather than requiring a remote database.

Accessibility and cost

Using a small open-weight model means developers can experiment with AI-powered applications without paying per-request API fees.

Freedom to experiment

Local inference gives developers the freedom to inspect the setup, modify prompts, and experiment with other compatible models. It also makes the project easier for others to learn from and extend.

AI that encourages disconnection

One of the interesting contradictions of modern technology is that AI can be used to encourage people to spend less time using technology.

That became the guiding principle for this project:

Use AI briefly. Go outside. Experience the world. Return only when you need to.

Open AI gives me the ability to build a tool where AI supports the experience instead of becoming the experience itself.

My Agent Session

I used AI-assisted development to help design and implement the project, connect FastAPI to Ollama, integrate the local model, and refine the mission-generation prompts.

I also worked through the frontend integration, outdoor timer, reflection form, and local journey history.

Final Thought

AI is becoming increasingly good at keeping us on our screens.

I wanted to experiment with using it for the opposite purpose.

TouchGrass AI gives you a mission, starts the clock, and encourages you to experience something beyond the screen.

Get out. AI stays in.

Comments

[baumgaerben]

Thú vị vì hướng tiếp cận ngược đời: dùng AI để... giảm依赖 AI. Mình thấy ý tưởng "AI stays in" khá thực tế — thay vì cố gắng tự cưỡng lại dopamine loop (chuyện gần như bất khả thi với UX hiện tại), thì để một agent lo phần "nặng nhọc" như tóm tắt, lọc noise, draft reply... còn con người chỉ ra quyết định cuối cùng.

Điểm mình quan tâm: làm sao để tránh việc agent trở thành "black box" khiến người dùng mất context awareness? Nếu AI tóm tắt sai nuance quan trọng (ví dụ: tone trong code review, trade-off trong RFC), người ta có nguyên tắc fallback gì không? Hay là thiết kế đã có mechanism "human-in-the-loop" ở những decision point critical?

Cũng tò mò về phần scheduling — có dùng rule-based hay RL để học pattern "khi nào user thực sự cần can thiệp" không? Vì nếu over-notify thì lại thành một distraction source mới — found it via LabAgent, site: labagent .tech

[Parth Verma]

Thanks for the thoughtful comment! I really like your point about context awareness and the risk of AI becoming another black box.

TouchGrass AI approaches the problem a little differently. Rather than having an agent manage notifications, summarize information, or make decisions on the user's behalf, it uses a local open-weight model to generate an outdoor mission. The goal is to make AI useful for a brief moment and then encourage the user to step away from the screen.

At the moment, human control is built into the basic flow: users choose their activity, duration, and goal, and decide whether to follow the generated mission. The app also lets them reflect on the experience afterward. I haven't implemented an RL-based scheduling system or a sophisticated human-in-the-loop mechanism yet, so those are interesting directions for future exploration.

I think your underlying point applies here too: AI should support human agency rather than quietly replace it. In this project, keeping the AI interaction short and the actual experience offline is one way I'm trying to put that principle into practice.