Skip to Content
ProductsResearch

Research

packages/research turns a tutorial into a reusable rules markdown file.

URL -> transcript -> LLM distill -> rules markdown

Basic run

bun packages/research/src/cli.ts \
  --url "https://www.youtube.com/watch?v=..." \
  --out rules.md

Offline proof

Use mock to prove extraction and output shape with no model.

bun packages/research/src/cli.ts \
  --url "<url>" \
  --provider mock \
  --out /tmp/rules.md

Already have a transcript

bun packages/research/src/cli.ts \
  --transcript-file transcript.txt \
  --title "Tutorial title" \
  --out rules.md

Transcript engine

The preferred path is yt-dlp because YouTube timed text can require visitor or PoToken handling.

brew install yt-dlp

Optionally point to a specific binary:

RESEARCH_YT_DLP=/path/to/yt-dlp bun packages/research/src/cli.ts --url "<url>"

Output rule

Research output should be concrete enough that an agent can build from it:

  • extracted mechanic or implementation rules
  • important constraints
  • reusable checklists
  • source URL and title
  • no secrets or personal data
AssetgenWarline