sudoleg / youtubegpt

YouTubeGPT is an LLM-based web-app that allows you to summarize and chat (Q&A) with YouTube videos.

YouTubeGPT- Your YouTube AI

Features ✨

YouTubeGPT lets you summarize and chat (Q&A) with YouTube videos. Its features include:

• provide a custom prompt for summaries ✍️ VIEW DEMO
  • you can tailor the summary to your needs by providing a custom prompt or just use the default summarization
• automatically save summaries 📂
  • the summaries can be automatically saved in the directory where you run the app. The summaries will be available under <YT-channel-name>/<video-title>.md
• get answers to questions about the video content ❓ VIEW DEMO
  • part of the application is designed and optimized specifically for question answering tasks (Q&A)
• create your own knowledge base 💾
  • once a video is processed, you can chat with it at any time!
• choose from different OpenAI models 🤖
  • currently available: gpt-3.5-turbo, gpt-4 (turbo), gpt-4o (mini)
  • by choosing a different model, you can summarize even longer videos and potentially get better responses
• experiment with settings ⚙️
  • adjust the temperature and top P of the model
• choose UI theme 🖌️
  • go to the three dots in the upper right corner, select settings and choose either light, dark or my aesthetic custom theme

Installation & usage

No matter how you choose to run the app, you will first need to get an OpenAI API-Key. This is very straightforward and free. Have a look at their instructions to get started.

build & run with Docker (or docker-compose)

1. make sure to provide an OpenAI API key (l. 43 in docker-compose.yml)
2. adjust the path to save the summaries (l. 39 in docker-compose.yml)
3. execute the following command:

# pull from docker hub
docker-compose up -d
# or build locally
docker-compose up --build -d

if you are only interested in summaries

# pull from Docker Hub
docker pull sudoleg/yotube-gpt:latest
# or build locally
docker build --tag=sudoleg/yotube-gpt:latest .
docker run -d -p 8501:8501 \
    -v $(pwd):/app/responses \
    -e OPENAI_API_KEY=<your-openai-api-key> \
    --name youtube-ai sudoleg/yotube-gpt:latest

ℹ️ For the best user-experience, you need to be in Tier 1 usage tier, which requires a one-time payment of 5$. However it's worth it, since then, you'll have access to all models and higher rate limits.

Contributing & Support 🤝

I’m working on adding more features and am open to feedback and contributions. Don't hesitate to create an issue or a pull request. Also, if you are enjoying the app or find it useful, please consider giving the repository a star ⭐

This is a small side-project and it's easy to get started! If you want to contribute, here’s the gist to get your changes rolling:

1. Fork & clone: Fork the repo and clone your fork to start.
2. Pick an issue or suggest One: Choose an open issue to work on, or suggest a new feature or bug fix by creating an issue for discussion.
3. Develop: Make your changes.
   • Ensure your code is clean and documented. Test the changes at least exploratively. Make sure to test 'edge cases'.
   • Commit your changes with clear, descriptive messages, using conventional commits.
4. Stay updated: Keep your branch in sync with the main branch to avoid merge conflicts.
5. Pull Request: Push your changes to your fork and submit a pull request (PR) to the main repository. Describe your changes and any relevant details.
6. Engage: Respond to feedback on your PR to finalize your contribution.

development in virtual environment

# create and activate a virtual environment
python -m venv .venv
source .venv/bin/activate
# install requirements
pip install -r requirements.txt
# you'll need an API key
export OPENAI_API_KEY=<your-openai-api-key>
# run chromadb (necessary for chat)
docker-compose up -d chromadb
# run app
streamlit run main.py

Technologies used

The project is built using some amazing libraries:

• The project uses YouTube Transcript API for fetching transcripts.
• LangChain is used to create a prompt, submit it to an LLM and process it's response.
• The UI is built using Streamlit.
• ChromaDB is used as a vector store for embeddings.

License

This project is licensed under the MIT License - see the LICENSE file for details.

youtubegpt's Issues

fix: deal with cases, where transcript in preferred language is translated

Here is an exanple of a video, where a transcript in prfererred language is available, but it's translated: https://youtu.be/J-HWmoTKhC8?si=bKz_6evS0pqouW8e

rn, an exception is thrown because there is neither a manual nor a autogenerated transcript (in en). however, a translated transcript in en is available

use openai python lib instead of requests

experiment with the temperature parameter

feat: add button for liking the video

setup renovate bot

ensure that context length is not exceeded

count tokens before submitting

feat: enable user to copy the summary to clipboard

• https://discuss.streamlit.io/t/how-to-add-copy-to-clipboard-button-to-text-area/49648/2
• https://discuss.streamlit.io/t/copy-to-clipboard-using-st-markdown/50415

feat: enable gpt-4o-mini

https://openai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence/

dockerizing

multi-arch build: https://www.docker.com/blog/multi-arch-build-and-images-the-simple-way/

test reproducible outputs

try to make outputs more deterministic

https://platform.openai.com/docs/guides/text-generation/reproducible-outputs

feat: let the user choose embedding model

implement configuration class

the config class should provide the ability to read config params from a file and adjust them according to the users preference

feat(chat): add option to display excerpts that were provided as context

setting whether to save summaries

retrieve completion in json format

to ensure that the model responses in specific format, json-mode can be used. this way i could retrieve the title of the summary and the summary itself as values from two different json keys.

Add transcription feature (separate page)

Most of the transcripts on YouTube are autogenerated. Thus, they contain grammatical errors, filler words, don't have any punctuation and structure. Maybe it can be fixed by using a speech-to-text model like OpenAI's whisper.

error when fetching metadata (title)

Video: https://www.youtube.com/live/BuMZzfDUOf8?si=JnMrM7pVJKexC1C5

Error:

ERROR:root:An unexpected error occurred: 'title'
Traceback (most recent call last):
  File "/app/pages/chat.py", line 214, in <module>
    video_metadata = get_video_metadata(url_input)
                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/modules/youtube.py", line 61, in get_video_metadata
    filename=f"{json_response['title']}",
                ~~~~~~~~~~~~~^^^^^^^^^
KeyError: 'title'
ERROR:root:An unexpected error occurred 'title'

open sourcing

• name for the project
• logo for the project
• installation/usage instructions
• #21
• License?
• contributing
• add overview on docker hub
• add section for known issues and limitations
• add screenshot or demo video

test: what happens when app is run wo API key?

feat: enable user to set preferred language

• preferred languages
• model
• temperature

handle potential errors

• missing transcript
• noEmbed svc unavailable
• errors wit video url

maintain context

in order to ask follow-up questions, it would be very useful to maintain context

• https://community.openai.com/t/how-to-maintain-context-with-gpt-3-5-turbo-api/94959

fix: collection name must start with alphanumeric character

https://youtu.be/_hi8qOZB2qU?si=dBJO13KM9d8UO-Mu

the cost should be estimated based on total number tokens

refactor: display link to Chat page only if chroma connections is established

Check API Key validity only at the start

rn the validity of the api key is checked every time the app/page is refreshed, which happens quite often due to streamlits's architecture. To improve performance, the validity of the api key should be checked only once

feat: add page for chat with an input field for video URL

feat(chat): add perplexity-like follow-up questions

Dependency Dashboard

This issue lists Renovate updates and detected dependencies. Read the Dependency Dashboard docs to learn more.

Open

These updates have all been created already. Click a checkbox below to force a retry/rebase of any.

• chore(deps): update dependency langchain-community to v0.3.1
• chore(deps): update dependency langchain-openai to v0.2.1
• chore(deps): update dependency langchain-text-splitters to v0.3.0
• chore(deps): update dependency pytubefix to v6.17.0
• chore(deps): update dependency openai-whisper to v20240927
• chore(deps): update dependency pytubefix to v7
• chore(deps): update dependency watchdog to v5
• Click on this checkbox to rebase all open PRs at once

Detected dependencies

docker-compose

docker-compose.yml

• chromadb/chroma 0.5.11

dockerfile

Dockerfile

• python 3.12.6-slim

github-actions

.github/workflows/docker.yml

• actions/checkout v4
• docker/setup-qemu-action v3
• docker/setup-buildx-action v3
• docker/login-action v3
• docker/build-push-action v6

.github/workflows/release.yml

• googleapis/release-please-action v4

.github/workflows/test.yaml

• actions/checkout v4
• actions/setup-python v5
• streamlit/streamlit-app-action v0.0.3

pip_requirements

requirements.txt

• youtube-transcript-api ==0.6.2
• langchain-community ==0.2.17
• langchain-openai ==0.1.25
• langchain-text-splitters ==0.2.4
• streamlit ==1.38.0
• watchdog ==4.0.2
• pytest ==8.3.3
• peewee ==3.17.6
• python-dotenv ==1.0.1
• chromadb ==0.5.11
• langchain-chroma ==0.1.4
• randomname ==0.2.1
• tiktoken ==0.7.0
• openai-whisper ==20231117
• pytubefix ==6.15.4

---

• Check this box to trigger a request for Renovate to run again on this repository

add options for desired output format

refactor(summary): create summaries for videos of any length

We can chunk up the transcript (perefreably using an advanced chunking method like semantic chunking), create summaries of the chunks and finally create a summary based on the "sub-summaries" of the chunks.

refactor: display available models by querying OpenAI's API