Personal Pi Agent

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

What I Built

I built a lightweight, always-on personal AI assistant powered by Google's Gemma open-weight model through OpenRouter and deployed on a DigitalOcean Droplet.

The assistant is accessible through Telegram, so I can use it directly from my phone without opening a separate AI app. It can summarize and triage emails, send emails, read and create calendar events, browse Google Drive, create folders, and upload documents or images received through Telegram.

The project was inspired by the idea of persistent personal assistants such as OpenAI's Dots and Meta's Muse. My goal was to build a smaller, practical version that could stay online continuously and interact with the tools I already use.

Code

Github: https://github.com/adityanagdekar/Lite-Weight-Pi-Agent

Acknowledgements

This project uses and extends the open-source pi-gateway project created by alejandro-ao.

I used pi-gateway as the Telegram-to-Pi gateway and modified it to support additional workflows such as Telegram document/image handling, attachment downloads, and follow-up instructions for uploaded files.

Full credit for the original pi-gateway implementation goes to its original author alejandro-ao.

How I Built It

The assistant runs on an Ubuntu DigitalOcean Droplet so it can remain available continuously without depending on my laptop.

Telegram acts as the user interface. Messages are received by pi-gateway, which maps each Telegram conversation to a persistent Pi session using SQLite.

Pi acts as the agent runtime and uses google/gemma-4-26b-a4b-it through OpenRouter for inference.

For actions involving Gmail, Google Calendar, and Google Drive, the agent uses the open-source Google Workspace CLI (gws).

I also modified pi-gateway to support Telegram documents and images. Attachments are downloaded to the agent workspace and their local file paths are passed to Pi, allowing the agent to act on files sent directly through Telegram.

Prize Categories

Best Use of DigitalOcean
Best Us of Gemma