Savify — AI Summaries for Reels, Videos & Blogs

Stop hoarding tabs and saved reels you'll never rewatch.

You know the drill: 47 open tabs, 200 saved Instagram reels, a YouTube "Watch Later" that's basically a graveyard. You saved all that knowledge — you just never get it back out.

This fixes that. Paste any URL — a blog post, a YouTube video, an Instagram reel — and in seconds you get a structured AI summary: key points, difficulty rating, tools mentioned, and the one takeaway that matters. Everything lands in a searchable personal knowledge base. On your phone? Just share the link to your Telegram bot and the summary comes right back in chat.

• Fast — summaries powered by gpt-oss-120b on Groq (fastest inference on the market)
• No captions? No problem — reels and caption-less videos get transcribed locally with Whisper
• Zero-friction mobile — share from Instagram to your Telegram bot: summarized, saved, done
• Yours — SQLite on your machine, your API keys, no third-party tracking

Built with: Python · FastAPI · Groq (gpt-oss-120b) · faster-whisper · SQLite · Vanilla JS

---

Screenshots

Homepage	Dashboard

---

Features

Universal Summarization

• Blogs & Articles — Scrapes any webpage with BeautifulSoup, strips the junk, summarizes the content.
• YouTube Videos — Fetches transcripts via the YouTube Transcript API. No transcript? Falls back to downloading the audio with yt-dlp and transcribing locally with faster-whisper.
• Instagram Reels — Downloads reel audio, transcribes with Whisper, summarizes.

Real-Time Progress

• Live SSE (Server-Sent Events) stepper shows every stage: detection, scraping/downloading, transcription, AI summary, saved.

Smart Knowledge Base

• AI Categorization — auto-tagged into AI, Web Dev, ML, Cybersecurity, or General.
• Difficulty Scoring — Beginner / Intermediate / Advanced.
• Smart Search — across titles, summaries, key points, and tools mentioned.
• Sorting & Filters — by date, difficulty, title, category, or source domain.

Personalization & Export

• Favorites — star summaries, filter to favorites only.
• Manual Editing — refine any AI summary; edits get an "Edited" badge.
• Export — copy as Markdown or download .md for Notion / Obsidian.

Telegram Bot

• Share from anywhere — send any URL to your bot, get the summary back in chat.
• Works locally out of the box — long-polling mode means no ngrok, no webhook setup for local dev.
• Same knowledge base — bot summaries appear on your dashboard too.

---

Getting Started

Prerequisites

Requirement	Details
Python	3.9+
ffmpeg	Audio processing. brew install ffmpeg (macOS) / sudo apt install ffmpeg (Linux)
Groq API Key	Free at console.groq.com/keys

1. Clone

git clone https://github.com/hemish22/savify.git
cd savify

2. Install dependencies

cd blog_summarizer/backend
pip install -r requirements.txt

faster-whisper downloads its model (~75 MB, tiny) automatically on first transcription.

3. Configure API keys

cp .env.example .env

Edit blog_summarizer/backend/.env:

GROQ_API_KEY=gsk_your_actual_key_here # required
TELEGRAM_BOT_TOKEN=123456:ABC-your_token_here # optional — only for the Telegram bot

.env is git-ignored. Your keys never leave your machine.

4. Run

uvicorn main:app --reload

5. Open

Page	URL
Homepage	http://localhost:8000
Dashboard	http://localhost:8000/dashboard

6. (Optional) Telegram bot — 2 minutes, no ngrok

1. Message @BotFather on Telegram, send /newbot, copy the token
2. Put it in .env as TELEGRAM_BOT_TOKEN
3. Restart the server — that's it

The server auto-detects local mode and uses long-polling: no public URL, no webhook, no tunnel. Send your bot any link and watch the summary come back. (When deployed to Render, it automatically switches to webhook mode instead.)

---

Deployment

Piece	Where	How
Backend	Render	render.yaml blueprint included — connect the repo, set GROQ_API_KEY + TELEGRAM_BOT_TOKEN in the dashboard. Telegram webhook registers itself on boot.
Frontend	Vercel	blog_summarizer/frontend/ deploys as a static site; it auto-points API calls at the Render backend when served from a *.vercel.app domain.

---

Project Structure

savify/
├── blog_summarizer/
│   ├── backend/
│   │   ├── main.py                 # FastAPI app, routes, SSE streaming, Telegram handling
│   │   ├── llm_service.py          # Groq (gpt-oss-120b) integration + robust JSON parsing
│   │   ├── scraper.py              # Blog/article scraping (BeautifulSoup)
│   │   ├── youtube_service.py      # YouTube transcripts + Whisper fallback
│   │   ├── instagram_service.py    # Instagram Reel detection + transcription
│   │   ├── audio_service.py        # yt-dlp audio downloading
│   │   ├── whisper_service.py      # faster-whisper transcription
│   │   ├── transcript_cleaner.py   # Filler-word removal, text cleanup
│   │   ├── telegram_service.py     # Bot API helpers: polling, webhook, formatting
│   │   ├── database.py             # SQLite schema + CRUD
│   │   ├── models.py               # Pydantic models
│   │   ├── requirements.txt
│   │   └── .env.example
│   ├── frontend/
│   │   ├── index.html              # Homepage — URL input + live progress
│   │   ├── dashboard.html          # Knowledge base UI
│   │   ├── styles.css              # Design system (dark mode)
│   │   └── script.js               # Filters, favorites, export, SSE client
│   └── docs/screenshots/
├── Dockerfile                      # Render deployment
├── render.yaml                     # Render blueprint
└── README.md

---

API Endpoints

Method	Endpoint	Description
GET	/	Homepage
GET	/dashboard	Dashboard
POST	/summarize-stream	Summarize with real-time SSE progress
POST	/summarize	Summarize (non-streaming)
GET	/summaries	All saved summaries
DELETE	/summaries/{id}	Delete a summary
POST	/summaries/{id}/favorite	Toggle favorite
PUT	/summaries/{id}/edit	Edit summary text
POST	/telegram-webhook	Telegram webhook (production mode)
GET	/telegram-setup	Manual webhook registration helper

Example

curl -X POST http://localhost:8000/summarize \
  -H "Content-Type: application/json" \
  -d '{"url": "https://example.com/some-article"}'

{
  "title": "Article Title",
  "domain": "example.com",
  "difficulty": "Intermediate",
  "category": "Web Dev",
  "summary": "A concise summary of the article...",
  "key_points": ["Key point 1", "Key point 2"],
  "takeaway": "The main actionable insight.",
  "original_url": "https://example.com/some-article"
}

---

Configuration Reference

All config lives in blog_summarizer/backend/.env:

Variable	Required	Description
GROQ_API_KEY	Yes	Free key from console.groq.com/keys
TELEGRAM_BOT_TOKEN	No	From @BotFather — only for Telegram integration

SQLite needs zero config. Whisper models download on first use.

---

Troubleshooting

Issue	Solution
ModuleNotFoundError	pip install -r requirements.txt
ffmpeg not found	brew install ffmpeg / sudo apt install ffmpeg
GROQ_API_KEY not set	Create .env (step 3)
Port 8000 in use	lsof -ti:8000 | xargs kill -9
Telegram bot silent	Another instance (e.g. a Render deploy) may own the webhook — the local server clears it on startup; restart the server
Instagram/YouTube download fails	pip install --upgrade yt-dlp

---

License

MIT — free for personal and commercial use.

---

Built by hemish22