Touch Grass AI: Offline AI That Turns Screen Time into Real-World Missions

Ishita chaudhary — Sat, 10 Oct 2026 19:02:49 +0000

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

What I Built

Touch Grass AI is an offline-first application designed to help people step away from their screens and engage with the real world.

Instead of recommending more content to consume, it generates personalized outdoor missions based on the user's mood, available time, energy level, budget, and surroundings.

For example, a user might say:

"I'm bored, have 30 minutes, and don't want to spend any money."

Touch Grass AI can generate a mission such as Mini Explorer, encouraging the user to explore a safe nearby area, discover different types of leaves, and notice things they usually overlook.

The goal is simple: use AI to encourage real-world experiences rather than more screen time.

Demo

🚧 Demo coming soon.

I'll add a screen recording or a live demo once the application is working.

Code

🚧 GitHub repository link coming soon.

How I Built It

The project is being developed using:

Python — application logic
Streamlit — user interface
Ollama — local model inference
Gemma — open-weight language model for generating personalized missions

The application is designed to run AI inference locally, without relying on a paid cloud AI API.
The mission-generation process takes the user's preferences and turns them into practical, time-bounded outdoor activities.

Why Does Open Innovation Matter?

Open innovation makes AI more accessible by allowing developers to experiment with open-weight models, local inference, and customizable tools.

For this project, running a model locally means that mission generation does not need to depend on a paid cloud AI API. It also provides an opportunity to explore more private, offline-first AI experiences.

I want to demonstrate that AI doesn't always need to keep us connected to a screen. It can help us disconnect and engage with the world around us.

My Agent Session

🚧 I'll add my DevRelay agent session or session link if available.

Prize Categories

I'll add the applicable prize categories after checking the official challenge requirements.

Built with 🌱 and the idea that sometimes the best thing technology can do is help us put it down.

CodeRecall - Explain My Own Code to Me(Local AI for a friend's Forgotten Repos)

Ishita chaudhary — Mon, 05 Oct 2026 06:35:47 +0000

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

My friend opens their own three-month-old repository and whispers: "why did I write this?" Comments are missing, commit messages say "fix stuff," and commercial AI assistants either give generic answers or require uploading the whole private repo to someone else's server.

CodeRecall is a local AI assistant that explains your own code back to you — not just what it does, but why it exists — using the repo itself as evidence: source code, tests, docs, and Git history.

Planned features (building this weekend):

📂 Index a local Git repo: code, README, tests, commits, diffs
💬 Ask things like "Why does this function exist?", "What breaks if I delete this?", "Explain this like I haven't seen it in a year"
🔍 Every answer cites file + line numbers and the commit that introduced the code (via git blame / git log)
🧠 Honest answers, split into three parts: Verified behavior → Historical evidence → Possible intent
🔌 100% offline. No API keys. No telemetry. Code never leaves the laptop.

Who it's for: my friend — and every developer who has ever been afraid of their own old repository. Bonus: I'll hand it over to them and share their reaction here.

Demo

In progress — I'm building this through the weekend and will update this section with a screen recording (running fully offline) plus my friend's reaction when I hand it over.

Code

GitHub repo:[https://github.com/ishitaachaudharyy/CodeRecall.git] — currently has the README, architecture, and build plan. Full implementation lands this weekend.

How I Built It

Everything runs locally

Gemma (Google's open-weight model) running locally via Ollama — the reasoning core
nomic-embed-text (local embeddings) for semantic search over the codebase
Tree-sitter to chunk code by function/class instead of random line counts
Chroma (local vector DB) + SQLite for metadata (file paths, commits, notes)
Gradio for a simple local UI
Plain git log / git blame / git show — no AI needed to extract the "why" evidence Pipeline: epo → parse (Tree-sitter) → chunk → embed locally → retrieve (code + docs + tests + commits) → Gemma → answer with citations

Why Does Open Innovation Matter?

Source code is the most sensitive thing a developer owns: credentials, client work, unreleased features. A closed AI product means uploading your private repo to a server you don't control — a non-starter for my friend's client work.

With open weights, CodeRecall:

Runs on my friend's laptop, fully offline — the code physically never leaves the machine
Costs $0 to run — no subscription, no per-token fees
Is swappable — Gemma 4B on their old laptop, Gemma 12B on mine; change one Ollama line
Is inspectable — we can read every prompt, tune the retrieval, and fix weird answers ourselves

A closed API literally cannot offer the core promise of this tool: your code stays yours.

My Agent Session

I'll link my DevRelay agent session here as I build this weekend.

Prize Categories

Best Use of Gemma — Gemma is the open-weight model at the heart of CodeRecall, running fully locally via Ollama with zero API calls.

DEV Community: Ishita chaudhary

Touch Grass AI: Offline AI That Turns Screen Time into Real-World Missions

What I Built

Demo

Code

How I Built It

Why Does Open Innovation Matter?

My Agent Session

Prize Categories

CodeRecall - Explain My Own Code to Me(Local AI for a friend's Forgotten Repos)

What I Built

Demo

Code

How I Built It

Why Does Open Innovation Matter?

My Agent Session

Prize Categories