Slow Signal：本機優先的 AI 待辦事項整理工具

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

What I Built

I built Slow Signal, a local-first brain dump sorter for the things I keep putting off. I paste in a messy list, and an open-weight AI model sorts it into what I could do today, what can wait, and what I can let go. It also suggests one tiny first step.

The idea is to make task sorting itself less of a task. No account, streak, or cloud AI service required.

Demo

https://pranjalkumar1307.github.io/StudySprint-AI/

To run the app locally, install Ollama and run ollama run gemma4:e2b, then start the Java server with mvn compile exec:java and open http://127.0.0.1:8787.

Code

https://github.com/PRANJALKUMAR1307/StudySprint-AI.git

How I Built It

Slow Signal has a small HTML, CSS, and JavaScript interface, with a Java 17 server connecting it to Ollama’s local API. The app lists locally installed models and asks the selected model to return a short, structured reflection: today, later, let go, and one first step.

The project is set up to use Gemma 4 through Ollama, and can also use other installed local models. The server filters out cloud models so brain dumps are sent to a local model.

Why Does Open Innovation Matter?

These notes can include personal details. Running an open-weight model locally means they don’t need to be sent to a hosted AI service. After the model is downloaded, the app can generate reflections on my computer without an internet connection.

Open models also let me switch models and change how the assistant sorts tasks. That makes the AI part something I can inspect and adapt to the way I want to use the app.

The Lazy Payoff

One brain dump replaces manually sorting every task into categories. The app is designed to return a short list and a single tiny first step, so I can spend less energy organizing and more energy deciding what I actually want to do.

I haven’t measured time saved yet, so I’m treating that as the goal rather than a benchmark.