Vidilearn extracts YouTube and web content locally for AI workflows
Vidilearn is an open-source that extracts YouTube transcripts, subtitles, chapters, articles, and structured metadata on a local computer. It does not require API keys, so it aims to reduce recurring costs for content collection. When modern JavaScript-heavy websites block simpler extraction tools, it can use Playwright as a fallback to open pages and collect content more reliably.
The extracted content can feed AI agents, , automation systems, Codex or Gemini CLI workflows, and stacks. Its features include YouTube transcript extraction, article cleaning, structured metadata, local embedding generation, and MCP server mode. Shared benchmark numbers include a 94.2% RAG hit rate, 92.1% precision, a 0.931 F1 score, and a 10.0 cost-adjusted score.
Claude is described as still slightly better in absolute accuracy, while Vidilearn gets close at near-zero operating cost. It can be installed with `npm i vidilearn`, and the is Alfo-Tech-Lab/vidilearn.
Key points
- Vidilearn is an open-source for extracting YouTube transcripts, articles, chapters, and metadata locally.
- It works without API keys, which can reduce recurring content ingestion costs.
- It is designed for AI agents, , automation systems, Codex and Gemini CLI workflows, and stacks.
- It includes a Playwright fallback for dynamic websites that simpler tools may fail to read.
- Shared benchmarks claim a 94.2% RAG hit rate, 92.1% precision, and a 0.931 F1 score.