TouchGrass — Ask AI. Hear the Plan. Touch the Grass.

Ryan Mehndiratta — Wed, 07 Oct 2026 15:54:21 +0000

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

What I Built

TouchGrass is a voice-first outdoor activity planner powered by open-source AI.
The idea is pretty simple: instead of using AI to keep people on their phones longer, I wanted to build something that uses AI to help people put their phones away.
The flow is:

Speak → Transcribe → Plan → Listen → Go outside

The user speaks their idea, ElevenLabs Scribe converts the speech into text, and a local open-weight AI model running through Ollama generates the outdoor plan.

The final plan can then be converted back into a natural voice briefing using ElevenLabs.

The goal isn't to create another AI assistant that you keep staring at.

Demo

Code

drunkPython056 / Touchgrass-ai

TouchGrass is a voice-first outdoor activity planner powered by open-weight AI.

TouchGrass 🌿

A local-first outdoor activity planner powered by open-weight AI.

Built for the Hacktoberfest Touch Grass challenge.

What it does

TouchGrass asks for a simple outdoor goal—walk, run, hike, garden or birding—then creates a short plan designed to get the user away from the screen quickly.

The important part: the AI runs locally through Ollama using an open-weight model. The app does not require a cloud AI API. If Ollama is unavailable, TouchGrass automatically falls back to a deterministic offline planner.

Why open innovation matters

Privacy: activity preferences and planning prompts can stay on the user's computer.
Offline resilience: the app has a no-model fallback and does not depend on a cloud AI service for the basic experience.
Model freedom: change OLLAMA_MODEL without changing the application code.
Low cost: local inference avoids per-request API charges.
Inspectable behavior: the prompt and server code are part of the project and can…

View on GitHub

How I Built It

For the AI planning layer, I use an open-weight Qwen model running locally through Ollama.

The model receives the user's request and creates an outdoor activity based on things such as:

Available time

Activity preference

Difficulty

Environment

Interests

Desired experience

Because the model runs locally, the core AI planning doesn't require sending the user's entire request to a cloud AI provider.

Why Does Open Innovation Matter?

For TouchGrass, open innovation isn't just about using an open model because the challenge asked for it.

It changes what the application can actually be.

With a closed AI API, the architecture would essentially become:

User → Cloud API → Response

With an open-weight model, I can instead build:

User → Local Model → Outdoor Mission

That gives developers much more control.

🔓 1. The AI can run locally

The planning model doesn't need to depend entirely on a remote AI service.

That makes the project more useful in situations where connectivity is limited — which is especially relevant for something whose purpose is to get people outdoors.

🧩 2. Models can be replaced

Qwen isn't permanently hardcoded into the concept.

A developer can experiment with another open-weight model depending on their hardware and requirements.

🔍 3. Developers can inspect and modify the system

Open models make experimentation possible.

Developers can change the prompts, agent logic, model, activity generation, and eventually even fine-tune the system for specific outdoor use cases.

🛠️ 4. Less vendor lock-in

The project doesn't need to be built around one proprietary AI provider.

ElevenLabs is used specifically where its technology adds value — voice transcription and voice generation — while the reasoning/planning layer remains open.

That separation is important.

Open AI handles the brain. ElevenLabs gives it a voice.

🌍 5. Open innovation makes experimentation easier

The most interesting part of open AI isn't simply getting a free model.

It's being able to take that model, combine it with other technologies, change the architecture, and build something that wasn't possible from a single closed product.

That's what I wanted to explore with TouchGrass.

My Agent Session

I used DevRelay during development to experiment with the project and its agent workflow.

Agent Session

[Add your DevRelay agent session link here]

The session can be used to show how the project was developed and how the AI-assisted workflow contributed to building TouchGrass.

DEV Community: Ryan Mehndiratta