<img src="docs/images/logo.png" alt="AskTube's Logo"/> AskTube - An AI-powered YouTube video summarizer and QA assistant powered by Retrieval Augmented Generation (RAG) 🤖 Run it entirely on your local machine with Ollama, or cloud-based models like Claude, OpenAI, Gemini, Mistral, and more

🏃🏽‍➡️ Demo & Screenshot

Watch "AskTube First Demo" on YouTube

https://github.com/user-attachments/assets/610ec00b-e25a-4ac5-900c-145c8485675f

💤 Features

[x] Work even with unsubtitle video
[x] No limit video time
[x] Support multiple AI vendors
[x] Focus on RAG implemetation
[x] Fully run on your local machine

🤷🏽 Why does this project exist?

I’ve seen several GitHub repositories offering AI-powered summaries for YouTube videos, but none include Q&A functionality.
I want to implement a more comprehensive solution while also gaining experience with AI to build my own RAG application.

🔨 Technology

Language: Python, JS
Server: Python@v3.10, Bun@v1
Framework/Lib: Sanic, Peewee, Pytubefix, Sentence Transformers, Sqlite, Chroma, NuxtJs/DaisyUI, etc.
Embedding Provider (Analysis Provider):
- [x] OpenAI
- [x] Gemini
- [x] VoyageAI
- [x] Mistral
- [x] Sentence Transformers (Local)
AI Provider:
- [x] OpenAI
- [x] Claude
- [x] Gemini
- [x] Mistral
- [x] Ollama (Local)
Speech To Text:
- [x] Faster-Whisper (Local)
- [x] AssemblyAI
- [x] OpenAI
- [x] Gemini

🗓️ Next Todo Tasks

[ ] Implement Speech To Text for cloud models
- [ ] AssemblyAI
- [ ] OpenAI
- [ ] Gemini
[ ] Enhance
- [x] ~Skip using RAG for short videos~
- [ ] Chat prompts, chat messages by context limit
- [ ] RAG: Implement Query Translation
  - [x] ~Multiquery~
  - [ ] Fusion
  - [ ] Decomposition
  - [ ] Step back
  - [ ] HyDE

🚀 How to run ?

For the first time running, the program maybe a bit slow due they need to install local models.

Run on your machine

Ensure you installed:
- Python 3.10
  - Windows User, please download here
  - Linux, MacOS User, please use homebrew or your install package command (apt, dnf, etc)
  - Or use conda
- Poetry
  - Windows User open Powershell and run:
```
(Invoke-WebRequest -Uri https://install.python-poetry.org -UseBasicParsing).Content | py -
```
  - Linux, MacOS User open Terminal and run:
```
curl -sSL https://install.python-poetry.org | python3 -
```
- Bun
- ffmpeg
  - MacOS User
```
brew install ffmpeg
```
  - Linux User
```
# Ubuntu
sudo apt install ffmpeg
# Fedora
sudo dnf install -y ffmpeg
```
  - Windows, please follow this tutorial Install ffmpeg for Windows

Clone repostiory

git clone https://github.com/jonaskahn/asktube.git

Create file .env in asktube/engine directory:
- Locally
- Free with some limitations
Run program
- You may need to run first:
```
poetry env use python
```
- Open terminal/cmd/powershell in asktube/engine directory, then run:
```
poetry install && poetry run python engine/server.py
```
- Open terminal/cmd/powershell in asktube/web directory, then run:
```
bun install && bun run dev
```
Open web: http://localhost:3000

With docker (In process)

Before You Start

I built these services to docker images, but if you want to build local images, please run build.local.bat for Windows or build.local.amd64.sh or build.local.aarch64.sh for MacOS, Linux

If you have a GPU (cuda or rocm), please refer ENV settings above, change params like above

Locally

Use local.yaml compose file to start
Open terminal/cmd/powershell in asktube directory

docker compose -f compose/local.yaml pull && docker compose -f compose/local.yaml up -d

After run, you need install Ollama model qwen2 and llama3.1 for QA

docker run ollama ollama run qwen2
docker run ollama ollama run llama3.1

Free (with rate limit)

You need to go Google Gemini and VoyageAI to register account and generate your own API keys:
- Gemini is free with your Google Account
- VoyageAI (recommended by Claude) gives you free 50M tokens (a huge amount) but you need to add your credit card first.
Replace your ENV setting in docker file free and start docker
Open terminal/cmd/powershell in asktube directory

docker compose -f compose/free.yaml pull && docker compose -f compose/free.yaml up -d

Ideal

Using VoyageAI for embedding texts
Using OpenAI and Claude for QA, register account and generate your own API keys
Replace your ENV setting in docker file ideal and start docker
Open terminal/cmd/powershell in asktube directory

docker compose -f compose/ideal.yaml pull && docker compose -f compose/ideal.yaml up -d

Result

Open web: http://localhost:8080

💡 Architecture

The real implementation might differ from this art due to its complexity.

1️⃣ Extract data from given URL

2️⃣ Storing embedding chapter subtitles

3️⃣ Asking (included enrich question)

🪧 Notice

Do not use this for production. This aimed for end-users on their local machines.

Do not request any advanced features for management.

Asktube

Install / Use

README