Sloth Digest: A Lazy Person’s Group Chat Helper Built with Gemma 4 on My Phone

#devchallenge #hf26challenge

Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady Submission 🦥

This is a submission for the Hacktoberfest Open-Source AI Challenge: Week 3

What I Built

Sloth Digest is a lazy person's helper. You paste a long group chat or message, and it gives you a 2-line summary, a short list of what you actually have to do, and three replies you can send: agree, say no politely, or ask a question. It also shows how many words you skipped reading.

It is for anyone who is in too many group chats, like class, office or society groups.

Demo

Try the app here: https://codepen.io/Nirwan-Sharma/full/MYpXYEN

See the code here: https://codepen.io/Nirwan-Sharma/pen/MYpXYEN

You need your own free Google AI Studio API key. The key stays in your browser.

Code

The app is one HTML file, all inside the CodePen. This is a part of the prompt it builds, which tells the model to use only what is pasted:

'Use ONLY what is written there.',
'Never invent names, dates, times, places or promises.',
'SUMMARY: at most 2 short lines.',
'TO DO FOR ME: short lines starting with a dash...',
'REPLIES: three short ready-to-send replies, numbered 1 to 3...'

How I Built It

I am a beginner and I built this on my phone, as one HTML file. It uses Gemma 4, an open-weight model from Google, through a free API key. A dropdown switches between Gemma 4 31B and the lighter 26B, and the app retries with the other model when the server fails. I used an AI assistant (Claude) to help write the code.

The hardest part was the API. It failed several times with server errors, empty answers and a key error. So I added a retry that swaps between the two models, and clearer error messages.

The Lazy Payoff

In my test I pasted a made-up group chat of 75 words. The app gave me back 52 words: a 2-line summary, a short to-do list and three replies. That is not a big saving for such a short chat, but it should save more on a long chat with hundreds of messages. Gemma got the key facts right (practice at 4, fees by the 25th, bring a white t-shirt), but it left out the Saturday photo shoot decision.

Why Does Open Innovation Matter?

Group chats are personal. The best version of this tool would run a model on my own computer, so my messages never leave it. Gemma is open-weight, so that is possible. In this first version I used Google's hosted API, so the pasted text does go to Google, and I say that in the app. I also added a warning to not paste private chats. Because the model is open, I could swap between two sizes with one dropdown, and later I can move to a local setup. For me, the open model is what makes that next step possible.

DEV Community