yt-summary

A tool to generate YouTube video summaries with LlmaIndex

The project was named yt-summary initially but PyPI rejected it for being too similar to an existing project. Hence it's renamed to yt-pls for the CLI tool (YouTube please)

Install

pip

pip install yt-pls

If you want to use the Gemini and/or Claude models, you need to install the optional dependencies:

pip install "yt-pls[google]"
pip install "yt-pls[anthropic]"

# or
pip install "yt-pls[all]"

Global

You can install the package globally with pipx or uvx:

pipx install yt-pls

uvx run yt-pls

Development

Install uv and run:

git clone https://github.com/nelnn/yt-summary.git
cd yt-summary
make install
source .venv/bin/activate

CLI

Export your API key(s) as an environment variable:

export OPENAI_API_KEY="sk-..."
export GOOGLE_API_KEY="..."
export ANTHROPIC_API_KEY="..."

Then execute

yt-pls [url/video_id]

You can specify the LLM provider and model:

yt-pls --provider openai --model gpt-5-mini-2025-08-07 [url/video_id]

You can also specify the summariser type:

yt-pls --summariser refined [url/video_id]

To save the output to a file, for example in markdown:

yt-pls [url/video_id] > summary.md

API

You can also use yt-summary as a library:

from yt_summary.run import get_youtube_summary

summary = await get_youtube_summary("https://www.youtube.com/watch?v=abc123")

Remember to export your LLM API key as an environment variable before running the function. Alternatively, you can save them in a .env file in your project root directory.

Transcript Extractor

To fetch the transcript and metadata:

from yt_summary.extractors import TranscriptExtractor

url = "url or video id"
transcript_extractor = TranscriptExtractor()
transcript_raw = await transcript_extractor.fetch(url)

This will return a Pydantic model YoutubeTranscriptRaw which looks like:

YoutubeTranscriptRaw(
    metadata=YoutubeMetadata(
        video_id="abc123",
        title="Video Title",
        author="Author Name",
        channel_id="channel123",
        video_url="https://www.youtube.com/watch?v=abc123",
        channel_url="https://www.youtube.com/channel/channel123",
        thumbnail_url="https://i.ytimg.com/vi/abc123/hqdefault.jpg",
        is_generated=True,
        language="English (auto-generated)",
        language_code="en",
    ),
    text="[00:00 (0s)] First sentence. Second sentence. [00:10 (10s)] Third sentence...",
)

Note: The extractor is just a wrapper around youtube-transcript-api which you can pass the same parameters to fetch as you would to YouTubeTranscriptApi.fetch but here we stitch the transcript snippets into a single string with timestamps embedded in the text. You can also pass the proxy settings to TranscriptExtractor as you would to YouTubeTranscriptApi.

Summariser

There are currently two summariser implementations: CompactSummariser and RefineSummariser.

CompactSummariser lists metadata, high level summary and Q&A with timestamps. It utilises the DocumentSummaryIndex from LLamaIndex.

RefineSummariser achieves the same by chunking the transcript to generate summaries for each chunk before consolidating them by making multiple calls to the LLM asynchronously. Be aware of the rate limits of your chosen LLM provider.

For example, to generate summary with CompactSummariser:

from yt_summary.extractors import TranscriptExtractor
from yt_summary.schemas import LLMModel, LLMProvidersEnum
from yt_summary.summarisers import CompactSummariser, RefinedSummariser

transcript_extractor = TranscriptExtractor()
transcript_raw = await transcript_extractor.fetch(url)
summariser = CompactSummariser(
    llm=LLMModel(
        provider=LLMProvidersEnum.OPENAI,
        model="gpt-5-mini-2025-08-07",
    )
)
summary = await summariser.summarise(transcript)

The repository is under development, expect breaking changes. Sugguestions and contributions are welcome!