Data

Tools

Indexes

Applications

Experiments

Awesome

An open API service indexing awesome lists of open source software.

https://github.com/cemroot/yt-ai-summarizer

Free AI-powered YouTube video summarizer — get summaries, key points & detailed analysis in 20+ languages. Supports Groq & Ollama Cloud. No subscription needed.
https://github.com/cemroot/yt-ai-summarizer

Last synced: 3 months ago
JSON representation

Free AI-powered YouTube video summarizer — get summaries, key points & detailed analysis in 20+ languages. Supports Groq & Ollama Cloud. No subscription needed.

Awesome Lists containing this project

README 

          
# YouTube AI Summarizer



> **Stop watching. Start reading.** Summarize any YouTube video for free with your own AI — no subscriptions, no hidden fees.



[![CI](https://github.com/CemRoot/yt-ai-summarizer/actions/workflows/ci.yml/badge.svg?branch=main)](https://github.com/CemRoot/yt-ai-summarizer/actions/workflows/ci.yml)

![Manifest Version](https://img.shields.io/badge/dynamic/json?url=https%3A%2F%2Fraw.githubusercontent.com%2FCemRoot%2Fyt-ai-summarizer%2Fmain%2Fmanifest.json&query=%24.manifest_version&label=manifest&prefix=v&color=blue)

![License](https://img.shields.io/badge/license-MIT-green)

![Version](https://img.shields.io/badge/dynamic/json?url=https%3A%2F%2Fraw.githubusercontent.com%2FCemRoot%2Fyt-ai-summarizer%2Fmain%2Fmanifest.json&query=%24.version&label=version&color=orange)



---



## What It Does



Turn any YouTube video into a concise, actionable summary in seconds. Bring your own AI key, choose your language, and get:



- **Summary** — Full picture in 3–5 clean paragraphs

- **Key Points** — 5–10 evidence-backed takeaways as a numbered list

- **Detailed Analysis** — Section-by-section breakdown with names, dates, data preserved

- **🎙️ AI Podcast** — Two-host NotebookLM-style audio conversation with Gemini TTS, volume control, and WAV download

- **💬 Chat** — Ask follow-up questions about the video and get AI answers grounded in the transcript



Summary, Key Points, and Detailed are generated in a **single API call** — no rate limit issues.



---



## Features



| Feature | Details |

|---------|---------|

| **Dual AI Provider** | Choose between **Groq** (ultra-fast) or **Ollama Cloud** (flexible models). Switch anytime. |

| **20+ Languages** | Output in English, Türkçe, Español, Français, Deutsch, 日本語, 한국어, 中文 and more — regardless of video language. |

| **Modern UI/UX** | Seamless YouTube integration, underline-style tabs, smooth animations, Inter font. |

| **Dark & Light Mode** | Automatically matches YouTube's theme + system preference. |

| **Onboarding Flow** | Step-by-step welcome page with provider selection and guided API key setup. |

| **First-Visit Tooltip** | Animated "Hey! I'm here" bubble guides new users to the button. |

| **One-Click Copy** | Copy any analysis to clipboard instantly. |

| **Auto-Summarize** | Optional: generate summaries automatically when you open a video. |

| **Smart Caching (LRU)** | In-memory + persistent storage with LRU eviction (max 20 videos). Tab switching is instant — even during active generation. |

| **Fun Facts on Loading** | 50 rotating "Did you know?" facts keep you entertained while AI processes. |

| **🎙️ AI Podcast** | NotebookLM-style two-host podcast with randomly paired male & female voices. Powered by Gemini TTS — volume control, WAV download for offline/WhatsApp sharing, completely free. |

| **💬 Video Chat** | Ask follow-up questions about the video. AI answers strictly from the transcript with full conversation history. |

| **🖥️ Fullscreen-Aware** | Extension UI auto-hides in fullscreen mode for distraction-free viewing. |

| **🌍 Multi-Language Onboarding** | Welcome page auto-detects browser language with manual selector. 11 languages: EN, TR, ES, FR, DE, JA, KO, ZH, PT, AR, HI. |

| **SPA-Compatible** | Works seamlessly with YouTube's single-page navigation. |

| **CI/CD Pipeline** | GitHub Actions: manifest validation, JS syntax checks, security audit, automated packaging. |



---



## Getting Started



### 1. Install



```bash

# Clone the repo

git clone https://github.com/CemRoot/yt-ai-summarizer.git



# Or download the latest release ZIP

```



Then in Chrome:

1. Navigate to `chrome://extensions/`

2. Enable **Developer mode** (toggle top-right)

3. Click **Load unpacked** → select the extension folder



### 2. Get a Free API Key



Choose your provider:



| Provider | Get Key | Free Tier |

|----------|---------|-----------|

| **Ollama Cloud** (recommended) | [ollama.com/settings/keys](https://ollama.com/settings/keys) | Gemini 3 Flash, 10+ models, free tier |

| **Groq** | [console.groq.com/keys](https://console.groq.com/keys) | 30 RPM, up to 500K tokens/day |



### 3. Configure & Go



1. Click the extension icon → paste your API key

2. Pick your AI model and output language

3. Open any YouTube video → click the floating brain icon

4. Done. Summaries in seconds.



---



## How It Works



```

YouTube Video → Extract Transcript → Send to AI Provider → Parse 3 Sections → Display

                                   ↳ Chat Mode → Transcript + Question + History → AI Answer

```



1. **Transcript Extraction** — Pulls caption data directly from YouTube using Android client context (no external APIs).

2. **Combined AI Call** — A single prompt generates Summary + Key Points + Detailed Analysis simultaneously.

3. **Robust Parsing** — 4-layer parser: exact delimiters → regex variants → heuristic headings → smart split.

4. **Instant Display** — Results shown in a native-feeling side panel. Switch tabs instantly from cache.

5. **Conversational Chat** — Chat tab sends questions alongside the transcript and conversation history for context-aware answers.



---



## AI Models



### Groq



| Model | Speed | Quality | Best For |

|-------|-------|---------|----------|

| Llama 3.3 70B | ⚡⚡ | ★★★★★ | Highest quality summaries |

| Llama 3.1 8B | ⚡⚡⚡ | ★★★ | Fast results, higher rate limit |

| Llama 4 Scout 17B | ⚡⚡ | ★★★★ | New generation model |

| Qwen3 32B | ⚡⚡ | ★★★★ | Strong reasoning |



### Ollama Cloud



| Model | Size | Best For |

|-------|------|----------|

| **Gemini 3 Flash** ⭐ | — | **Recommended** — Fast & high quality |

| Qwen3-Next 80B | 80B | Reasoning & thinking |

| DeepSeek V3.2 | 671B | Most powerful |

| GPT-OSS 120B | 120B | General purpose |

| Kimi K2.5 | — | Multimodal |

| Devstral Small | 24B | Code-focused |



---



## Project Structure



```

yt-ai-summarizer/

├── manifest.json              # Extension manifest (V3)

├── service-worker.js          # Background: AI calls, routing, parsing

├── content/

│   ├── content.js             # Main controller (cache, SPA nav, chat)

│   ├── content.css            # Panel styles + dark/light theme + chat UI

│   ├── transcript.js          # YouTube transcript extraction

│   ├── ui.js                  # Panel UI, tabs, chat interface, onboarding tooltip

│   ├── podcast.js             # Podcast audio player

│   └── page-bridge.js         # MAIN world bridge for YT data

├── popup/

│   ├── popup.html             # Settings panel (dual provider)

│   ├── popup.js               # Settings logic

│   └── popup.css              # Modern card-based styles

├── update/

│   ├── update.html            # What's New page shown after extension updates

│   ├── update.js              # Changelog rendering + i18n + actions

│   └── update.css             # What's New page styles

├── welcome/

│   ├── welcome.html           # Step-by-step onboarding

│   ├── welcome.js             # Provider selection & validation

│   └── welcome.css            # Onboarding styles

├── utils/

│   ├── storage.js             # Provider-aware storage helpers

│   └── gemini-pcm-wav.js      # Gemini TTS PCM→WAV export + download

├── icons/                     # Extension icons (16, 48, 128px)

├── _locales/

│   ├── en/messages.json       # English strings

│   └── tr/messages.json       # Turkish strings

├── privacy-policy.html        # Privacy policy (bundled with extension)

├── privacy-policy.js          # TOC highlight script for privacy page

├── docs/                      # GitHub Pages (uninstall feedback, etc.)

│   ├── index.html

│   └── uninstall.html

├── .github/workflows/

│   ├── ci.yml                 # CI/CD pipeline

│   └── github-pages.yml       # Deploy docs/ to GitHub Pages

└── README.md

```



---



## CI/CD Pipeline



Every push to `main` triggers 5 automated checks:



| Job | What It Does |

|-----|-------------|

| **Manifest Check** | Validates JSON, required fields, manifest_version 3, referenced file existence |

| **JS Lint** | Syntax checks on all `.js` files, JSON validation, HTML structure checks |

| **Security Audit** | Scans for hardcoded API keys, `eval()`, `document.write()`, CSP validation, host permission audit |

| **Version Consistency** | Ensures version numbers match across manifest.json, privacy-policy.html, and README.md |

| **Build & Package** | Creates versioned `.zip` artifact for Chrome Web Store submission |



### Automated Maintenance



| Workflow | Schedule | What It Does |

|----------|----------|-------------|

| **YouTube Version Monitor** | Mon & Thu 09:00 UTC | Compares ANDROID client version against NewPipeExtractor, auto-creates PR if outdated |

| **Transcript Health Check** | Daily 06:00 UTC | Tests InnerTube caption extraction against a known video, opens GitHub Issue on failure |



---



## Privacy



- **Zero data collection** — no analytics, no tracking, no telemetry

- **Local storage only** — API keys and settings stay in `chrome.storage.local` on your device

- **Direct API calls** — transcripts and chat prompts go only to the AI hosts you enabled (Groq, Ollama Cloud, Gemini TTS)

- **Permissions** — `storage`, `activeTab`, plus declared host access for YouTube and those providers (see `manifest.json`)

- **Open source** — inspect every line of code



See the full [Privacy Policy](privacy-policy.html).



### Uninstall feedback URL (custom domain)



After uninstall, Chrome opens a page you host over **HTTPS** (required by `chrome.runtime.setUninstallURL`). This project uses your portfolio domain:



**https://cemkoyluoglu.codes/yt-ai-summarizer/uninstall.html**



The constant `UNINSTALL_FEEDBACK_URL` in `service-worker.js` must match that URL exactly (including `www` vs apex if you standardize on one).



**Deploy on Vercel (portfolio site):**



1. In this repo, the canonical HTML is [`docs/uninstall.html`](docs/uninstall.html).

2. In your **portfolio** Vercel project, add the same file at **`public/yt-ai-summarizer/uninstall.html`** (Next.js, Vite, or any setup that serves `public/` at the site root).

3. Deploy. Confirm the URL loads in an incognito window.

4. Whenever you change `docs/uninstall.html` here, copy it to the portfolio repo again and redeploy — then ship a new extension version if the URL path ever changes.



**Why not `github.io`?** Users only see the domain above; `cemroot.github.io` is no longer used for uninstall.



**Optional — GitHub Pages backup:** [`.github/workflows/github-pages.yml`](.github/workflows/github-pages.yml) can still publish `docs/` for your own testing or redirects; the extension does not point there unless you change `UNINSTALL_FEEDBACK_URL` back.



---



## Requirements



- Chrome 111 or later

- A free API key from [Groq](https://console.groq.com/keys) or [Ollama](https://ollama.com/settings/keys)



---



## Pre-release checklist (maintainers)



- **Security:** CI runs secret-pattern grep, CSP check, and manifest file references — keep `host_permissions` minimal.

- **Debug noise:** No `console.log` / `console.warn` / `debugger` in shipping paths; `console.error` is only used for serious failures (e.g. missing content-script dependencies).

- **Architecture:** Prefer ES classes and shared helpers (`StorageHelper`, controllers, UI modules) — avoid duplicating secrets or API base URLs outside `utils/storage.js` / `service-worker.js`.

- **Versions:** `manifest.json`, `privacy-policy.html`, and the top README changelog entry must stay aligned (enforced by the **Version Consistency** CI job).

- **Uninstall page:** After editing `docs/uninstall.html`, copy it to the portfolio Vercel project at `public/yt-ai-summarizer/uninstall.html` and redeploy so it matches `UNINSTALL_FEEDBACK_URL`.



---



## Troubleshooting



### "This extension is not trusted by Enhanced Safe Browsing"



This warning appears for **all new extensions** on the Chrome Web Store. It is **not** a security issue — it simply means the developer account hasn't yet built a long enough compliance history with Google. The warning disappears automatically after a few months. You can safely click **"Continue to install"** to proceed.



### CRX_FILE_NOT_READABLE on reinstall



If you uninstall the extension and immediately try to reinstall from the Chrome Web Store, Chrome may show a `CRX_FILE_NOT_READABLE` error. This is caused by Chrome's internal download cache holding a stale reference.



**Fix:** Close Chrome completely, reopen it, then install the extension again from the Web Store. This clears Chrome's session cache and allows a fresh download.



### Extension not appearing on YouTube



- Make sure you're on `youtube.com` (not an embedded player)

- Check that the extension is enabled at `chrome://extensions/`

- Try refreshing the YouTube page (Ctrl+R / Cmd+R)

- If you just installed, wait a moment and navigate to a new video



---



## Changelog



### v1.8.2



- **🎨 Update page polish**: Refined `update/update.css` for a more professional What's New layout and visual hierarchy. Fixed top accent line styling to match the welcome page.

- **✨ Hero icon update**: Replaced the oversized red star SVG with the actual extension icon (`icon128.png`) in the update page's hero section.

- **🔖 YouTube favicon**: Added a YouTube-style SVG favicon and wired it to update + popup pages.

- **🧭 Open Settings behavior**: What's New now requests opening the extension action popup first, with an elegant in-page Settings Toast fallback if the popup API fails, instead of opening an ugly new tab.

- **📜 Previous Versions toggle**: Fixed a CSS conflict (`display: flex` overriding `hidden`) making the previous versions list permanently visible instead of hidden by default.



### v1.8.1



- **📦 Store resubmit**: Bumped manifest version above 1.8.0 so Chrome Web Store accepts a new package (same codebase as 1.8.0 feature set).

- **🏷️ Version label**: Popup and What’s New read `chrome.runtime.getManifest().version` — no hardcoded version string.

- **📁 Package**: Distribution ZIP includes `update/` (What’s New page assets).



### v1.8.0



- **🎙️ Podcast TTS Model Fix**: Migrated from `gemini-2.5-flash-tts` (non-existent in v1beta) to `gemini-2.5-flash-preview-tts` — the correct model ID per Google's official speech generation docs. Fixes "model not found" error that broke all podcast generation.

- **🔊 Podcast Volume Control**: New volume slider (0–100%) on the podcast player. Setting persists across videos via `chrome.storage.local`.

- **📥 Podcast WAV Download**: "Download audio (WAV)" button exports the Gemini TTS audio as a proper WAV file. Users can save locally or share via WhatsApp as a document for offline listening.

- **🔄 Non-Blocking Tab Switching**: Completely reworked tab switch architecture — each pipeline (summary, podcast, chat) now has its own busy flag. Switching tabs while AI is generating no longer freezes the UI or shows stale content from another tab. Cache-first display ensures instant tab switches.

- **🧹 Clean Content Transitions**: `switchMode` now clears the content area synchronously before dispatching async handlers, eliminating brief flashes of wrong-tab content.



### v1.7.2



- **🔗 Uninstall URL**: Post-uninstall page now opens on **https://cemkoyluoglu.codes/yt-ai-summarizer/uninstall.html** (portfolio / Vercel) instead of GitHub Pages — same markup as `docs/uninstall.html`; host that file under `public/yt-ai-summarizer/` on Vercel.



### v1.7.1



- **📄 Privacy policy**: Redesigned layout and copy (Chat, Gemini TTS, storage, permissions). Last updated date refreshed.

- **🌐 GitHub Pages**: `docs/index.html` landing polish; uninstall page uses an inline SVG favicon (no missing `favicon.png` on Pages).

- **📖 README**: Project structure, Pages/uninstall URL, optional custom-domain notes, maintainer pre-release checklist.

- **🔧 Version sync**: Patch bump so store/build artifacts match the updated bundled `privacy-policy.html`.



### v1.7.0



- **💬 Interactive Video Chat**: New "Chat" tab lets you ask follow-up questions about the video. AI answers strictly from the transcript with full conversation history (last 10 messages). Works with all providers (Groq, Ollama, Gemini).

- **🖥️ Fullscreen Auto-Hide**: Extension toggle button and panel automatically hide when YouTube enters fullscreen mode — no more UI clutter during cinema viewing.

- **🧹 Pre-Release Cleanup**: Removed routine `console.log` / `console.warn` debugging from production paths; `console.error` remains only for serious failures (e.g. missing dependencies).

- **🐛 Podcast Rate Bug Fix**: Fixed a race condition in `podcast.js` where changing playback speed caused a small seek jump. Position is now captured before the rate changes.

- **📏 Chat Context Alignment**: Chat transcript context window set to 80K characters, matching the summary pipeline.



### v1.6.4



- **Gemini TTS Model Update**: Updated TTS model identifier.

- **Auto-Merge Pipeline**: Version Monitor PRs now auto-merge via squash after all CI checks pass — zero manual intervention needed.

- **Branch Protection**: All PRs require 5 CI checks before merging (Manifest, Lint, Security, Version Consistency, Build).

- **CODEOWNERS**: Critical files auto-assign @CemRoot as reviewer on PRs.



### v1.6.3



- **Fix Empty Transcript**: Updated YouTube InnerTube ANDROID client from v19.29.37 to v21.03.36 (SDK 35, Android 15). YouTube was rejecting the outdated client version and returning empty captions.

- **Dynamic Client Version**: Extension now auto-extracts YouTube's current WEB client version at runtime as a fallback when the ANDROID client fails. No more manual version updates for WEB client changes.

- **YouTube Version Monitor**: New GitHub Actions workflow checks NewPipeExtractor twice a week and auto-creates PRs when the ANDROID client version is outdated.

- **Transcript Health Check**: Daily automated probe tests InnerTube caption extraction and opens a GitHub Issue if transcript fetching is broken.

- **Version Consistency CI**: New CI job ensures version numbers stay in sync across manifest.json, privacy-policy.html, and README.md.

- **Uninstall URL**: Added `uninstall_url` to manifest for post-uninstall feedback and reinstall guidance.

- **Troubleshooting Guide**: Added FAQ section to README covering Enhanced Safe Browsing warning, CRX_FILE_NOT_READABLE reinstall issue, and common setup questions.



### v1.6.2



- **Version Bump**: Force Chrome Web Store CRX regeneration to resolve CDN caching issues.



### v1.6.1



- **🎛️ New Language Selector**: Searchable, two-column language menu with flags and modern "pill" trigger on the welcome page.

- **💾 Cache Controls**: Added "Cache summaries" and "Cache transcripts" toggles to settings; transcript caching disabled by default, reducing RAM/disk usage.

- **🔐 API Key Obfuscation**: Keys stored with XOR + Base64 encoding in chrome.storage instead of plaintext.

- **🔧 Aligned Defaults**: Provider and model defaults now match across all entry points — Ollama + Gemini 3 Flash.

- **🗑️ Quick Cache Clear**: One-click trash icon in the panel header to clear cached data.

- **🔒 CI Security Checks**: Obfuscation and sender.id validation checks added to the CI pipeline.

- **🌍 Welcome Page i18n Fix**: Fixed a bug where instruction steps, Validate button, and footer stayed in English when switching languages.



### v1.6.0



- **🎭 Random Voice Pairs**: Podcast voices are no longer always male-male. 10 voice pair combinations with male-female, female-male, and female-female options randomly selected per video. Uses Gemini TTS voices: Charon, Kore, Puck, Aoede, Fenrir, Leda, Orus, Zephyr.

- **🌍 Multi-Language Welcome Page**: Onboarding page now supports 11 languages (English, Türkçe, Español, Français, Deutsch, 日本語, 한국어, 中文, Português, العربية, हिन्दी). Auto-detects browser language with manual language selector. RTL support for Arabic.

- **📋 Inline Gemini API Key**: Removed "Open Settings" redirect. Users can now enter their Gemini API key directly inside the podcast panel without leaving the page.

- **🔒 Security Audit**: Comprehensive code review — no XSS vectors, API keys stored safely in chrome.storage.local, all user input sanitized with escapeHtml, CSP enforced, no eval/document.write usage.

- **📄 Privacy Policy**: Updated to include Google Gemini TTS data flow and host permission documentation.

- **Step 3 updated**: "How It Works" now mentions the Podcast mode alongside Summary, Key Points, and Detailed Analysis.



### v1.5.0



- **🎙️ Gemini TTS Podcast**: Replaced Web Speech API with Google Gemini 2.5 Flash TTS

  - Near-human quality voices (Charon for Alex, Puck for Sam)

  - Multi-speaker audio generated as a single audio file

  - Real audio player: play/pause, seek bar, 10s skip, speed control (0.75x–1.5x)

  - Audio duration displayed, progress bar click-to-seek

- **Gemini API Key**: New settings field for free Gemini API key (no credit card needed)

- **Guided setup**: Step-by-step instructions shown when Gemini key is missing

- **Region restrictions**: Friendly "not available" messages for unsupported regions (China, Iran, Russia, North Korea) in their native languages

- **Error handling**: Specific messages for GEMINI_REGION_BLOCKED, GEMINI_RATE_LIMITED, GEMINI_KEY_MISSING



### v1.4.0



- **🎙️ AI Podcast**: NotebookLM-style two-host podcast generated from video summaries

  - AI writes a natural conversation script between two hosts (Alex & Sam)

  - Web Speech API plays it with two distinct voices — completely free, no extra API needed

  - Full podcast player: play/pause, skip forward/back, speed control (0.75x–1.5x)

  - Live subtitles showing who's speaking and what they're saying

  - Scrollable transcript with click-to-jump navigation

  - Podcast tab added alongside Summary, Key Points, and Detailed Analysis

- New `podcast.js` engine: voice selection, state management, sequential speech playback



### v1.3.0



- **Fun Facts**: 50 rotating "Did you know?" facts displayed during AI processing to keep users entertained

- **Default provider changed**: Ollama Cloud is now the recommended provider (was Groq)

- **Recommended model**: Gemini 3 Flash set as default for Ollama Cloud

- Welcome page, popup, and storage defaults all updated to reflect Ollama Cloud + Gemini 3 Flash

- Facts rotate every 5 seconds with smooth fade animation



### v1.2.3



- Performance: ~3x faster for long videos (chunk size 24K → 80K, parallel processing)

- Most 1-hour videos now fit in a single chunk — no pre-summarization needed

- Remaining chunks processed in parallel batches of 3 via `Promise.all()`



### v1.2.2



- LRU cache algorithm: max 20 videos cached, oldest auto-evicted when limit is reached

- Each video stores up to 4 keys (3 summaries + 1 transcript), max ~800 KB total

- Cache reads update `lastAccessed` timestamp — frequently watched videos stay cached longer

- `clearCache()` now also wipes the LRU index



### v1.2.1



- Fixed: Toggle button not appearing on SPA navigation (broadened content script matching)

- Added: "Summarize this video?" start prompt — no API call until user clicks Start (saves credits)

- Localized start prompt in 8 languages (EN, TR, ES, FR, DE, JA, KO, ZH)

- Fixed: CI false positive on placeholder API keys



### v1.2.0



- Dual AI provider support (Groq + Ollama Cloud)

- Complete UI/UX redesign (Inter font, card-based settings, underline tabs)

- 20+ language support with flag emojis in selector

- First-visit onboarding tooltip ("Hey! I'm here")

- Robust 4-layer response parser (fixes all-tabs-same-content bug)

- GitHub Actions CI/CD pipeline

- Updated privacy policy for both providers

- Marketing-focused Chrome Web Store descriptions



### v1.0.2



- Fix: credentials added to all YouTube fetch calls



### v1.0.1



- Fix: empty transcript bug in transcript extraction



### v1.0.0



- Initial release



---



## Contributing



1. Fork the repo

2. Create a feature branch (`git checkout -b feature/amazing`)

3. Commit your changes

4. Push and open a Pull Request



---



## License



MIT License — free to use, modify, and distribute.



---





  Built for people who learn faster by reading.


  Built by people who got tired of 45-minute videos with 3 minutes of useful content.