Deploy

Deploy

Deploys the joost-blog site to Cloudflare Pages.

What it does

1. Fetches YouTube transcripts for any new videos (scripts/fetch-transcripts.mjs)
   • Uses cached transcripts when the file already exists
   • Requires yt-dlp on PATH (install via brew install yt-dlp)
2. Regenerates the NLWeb search index (scripts/generate-nlweb-index.mjs)
   • Scans all blog posts, pages, and videos
   • Skips draft content
   • Appends each video's transcript to its search body
   • Generates embeddings for new/changed content via Cloudflare API
   • Uses cached embeddings for unchanged content
3. Stages the updated index + any new transcript files
4. Commits with a descriptive message
5. Pushes to origin, triggering the Cloudflare Pages deploy

Steps

1. Run node scripts/fetch-transcripts.mjs and show the output to the user
2. Run node --env-file-if-exists=.env scripts/generate-nlweb-index.mjs and show the output
3. Check what changed under src/generated/ with git status --short src/generated/
4. Stage: git add src/generated/nlweb-index.json src/generated/nlweb-index.mjs src/generated/embedding-cache.json src/generated/transcripts/ (plus any other staged content the user is shipping)
5. Run git status to show the user what will be deployed
6. Ask the user for a commit message (suggest one based on what changed)
7. Commit and push to origin
8. Confirm the deploy was triggered

Notes

• The CF_API_TOKEN and CF_ACCOUNT_ID environment variables must be available for embedding generation; they live in .env. Pass --env-file-if-exists=.env to Node so the script picks them up.
• The prebuild step was removed from package.json — the index must be regenerated locally before deploying.
• If only code changes (no content changes), the index regeneration will show "0 to generate" and the cached index is reused.
• fetch-transcripts.mjs is a no-op for videos that already have a cached transcript, so it's safe to run on every deploy.