ytq 0.1

Build knowledge base from YouTube video transcripts

ytq

Build knowledge base from YouTube video transcripts

Overview

ytq (short for YouTube Query) is a CLI tool that processes YouTube videos to create a searchable knowledge base. It:

• Downloads and extracts transcripts from YouTube videos
• Uses LLMs to generate structured summaries
• Each transcript is split into multiple chunks (subsections). Each section preserves its start and end times timestemps. Then chunks are embedded using openai 'text-embedding-3-small' model. Created embeddings are used when --semantic search flag is enabled.
• Stores everything in a searchable SQLite database
• Provides a CLI for adding videos, searching, and viewing summaries

Installation

Install this tool using pip:

pip install ytq

If you are using uv then you can run directly the cli in temporary enviironment like so:

uvx ytq <command> <args>

or you can also install it as a tool:

uv tool install ytq
# and then
ytq <command> <args>

Usage

Adding a Video to the Knowledge Base

To add a YouTube video to your knowledge base, use the add command:

ytq add <video_url>

Optional parameters:

• --chunk-size: Maximum size of each text chunk (default: 1000 characters)
• --chunk-overlap: Overlap between chunks (default: 100 characters)
• --provider: LLM summarization provider (default: "openai")
• --model: LLM summarization model (default: "gpt-4o-mini")

Example:

ytq add https://youtube.com/watch?v=example --chunk-size 1500 --provider anthropic

If you try storing a video that is already in the db, the old version is removed and replaced with the new version.

Searching the Knowledge Base

Search your knowledge base using the query command:

ytq query <search_term>

Search options:

• --chunks: Enable chunk-level search
• --semantic: Enable semantic search (when chunk search is enabled)
• --limit: Maximum number of results (default: 3)

Examples:

# Video-level search (default)
ytq query "machine learning"

# Chunk-level keyword search
ytq query "neural networks" --chunks

# Semantic chunk-level search
ytq query "types of algorithms" --chunks --semantic

Viewing Video Summary

To view a summary of a specific video:

ytq summary <video_id>

Example:

ytq summary dQw4w9WgXcQ

Deleting a Video

To remove a video from the knowledge base:

ytq delete <video_id>

Version Information

To check the version of ytq:

ytq --version

Development

To contribute to this tool, first checkout the code. Then create a new virtual environment:

cd ytq
python -m venv venv
source venv/bin/activate

Now install the dependencies and test dependencies:

pip install -e '.[test]'

To run the tests:

python -m pytest