GentleBrief：以開放權重 AI 製作每日音訊摘要，減少無止境刷屏

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

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

What I Built

Modern web browsing often feels like cognitive overload. Checking the morning news across Hacker News, tech blogs, and research hubs frequently pulls readers into endless doomscrolling—fighting clickbait headlines, sensationalist takes, cookie banners, tracking scripts, and algorithmic rage-bait.

GentleBrief acts as a mindful anti-doomscroll barrier.

Instead of opening 40 tabs and exhausting your mental bandwidth before breakfast, GentleBrief automates the morning check-in:

You pick the desks that matter to you (or drop in any public blog/RSS link).
It quietly sends a cloud browser to extract the articles behind the scenes.
An open-weight model strips away the hysteria, point tallies, and clickbait, distilling the developments into three calm, grounded facts.
It narrates the digest into a conversational, ambient audio briefing that you can listen to hands-free while getting ready, walking, or drinking your morning coffee.

It lets developers and knowledge workers be intentionally "lazy" about consuming information—staying fully informed without ever getting sucked into a screen.

Demo

Live Application: gentle-brief.onrender.com
Features in action:
Automated cloud extraction bypassing tracking scripts and popups.
Custom calm-pacing voice narration with soothing ambient soundscapes.
Per-user client-side isolated playback history.
Live anti-doomscroll metrics highlighting filtered headlines and minutes of screen time saved.

Code

Gentle Brief

Gentle Brief is a Next.js app for turning a curated set of sources into a polished daily brief. It reads the latest input from selected feeds, distills the signal, and presents a readable summary with pacing and voice controls.

What it does

Selects and combines topic sources from built-in and custom URLs
Brews a short digest from those sources
Supports a pacing menu and voice ambience settings
Keeps a history of generated digests for replay
Uses a browser-based or live pipeline depending on configuration

Tech stack

Next.js 16
React 19
TypeScript
Tailwind CSS
Sonner for toast notifications

Getting started

Install dependencies:
```
npm install
```
Start the development server:
```
npm run dev
```
Open the app in your browser:
```
http://localhost:3000
```

Available scripts

npm run dev
npm run build
npm run start
npm run lint

Project notes

This project is a front-end digest dashboard with a generated brief workflow. The main app…

Built with:

Framework: Next.js (App Router, TypeScript)
Styling & UI: Tailwind CSS, Radix UI, shadcn/ui
Cloud Scraping: Browserbase SDK
Inference Engine: Groq API running open-weight Meta Llama 3 / Google Gemma
Audio Synthesis: ElevenLabs Conversational Voice API

How I Built It

The pipeline is built around an open-weight inference architecture orchestrated through Next.js:

Headless Extraction (Browserbase): Rather than fragile client fetches that fail against bot walls and modern SPAs, GentleBrief spins up headless cloud browser sessions via Browserbase. It navigates to configured desks (e.g., Hacker News, Indie Hackers, AP News, Phys.org) and scrapes strictly semantic article content, intentionally discarding navbars, paywall banners, and ad clutter.
Open-Weight Distillation (Llama 3 / Gemma on Groq): The raw scrape payload is routed to open-weight LLMs via Groq. The system prompt is fine-tuned to function as a stoic, mindful editor. It is explicitly instructed to eliminate hype, ignore upvote/comment counts, reject fear-mongering framing, and synthesize the day's core signals into peaceful, clear sentences formatted for vocal narration.
Ambient Voice Narration (ElevenLabs): The distilled script is sent to the ElevenLabs API using voice profiles like Charlotte (soft, gentle) or Adam (calm, conversational). The generated audio stream is fed to the dashboard player with custom scrubbing, speed controls, and optional ambient sound layers.

Why Does Open Innovation Matter?

Open-source and open-weight AI (like Llama 3 and Gemma) are essential for personal software like GentleBrief.

Information filtering is inherently subjective. When news curation is locked behind closed, proprietary platforms, users are held hostage by engagement algorithms designed to provoke anxiety and maximize screen time. Open models give developers full control over the editorial ethos: we can inspect the prompts, enforce complete neutrality, self-host or change inference providers at will, and guarantee that our tools serve human mental well-being rather than advertising metrics.

Open innovation makes calm technology accessible without closed-ecosystem gatekeeping.

The Lazy Payoff

GentleBrief replaced a fragmented morning habit:

Before: 30–45 minutes bouncing across Hacker News, Twitter, and tech sites, leaving multiple tabs open and feeling mentally scattered before work.
With GentleBrief: 0 minutes of doomscrolling. Clicking Brew Digest runs the cloud pipeline in seconds, producing a 60-to-90-second audio recap with three clear bullet points.
Estimated impact: Saves roughly ~25 minutes of screen fatigue every single morning.

Prize Categories

Best Use of ElevenLabs: Dynamic voice synthesis generating tranquil, ambient audio dispatches.
Best Use of Browserbase: Headless cloud browsing automating the extraction of public news feeds and custom URLs without ads or tracking.