A toolchain that periodically fetches customer reviews from Google, Facebook and Trustpilot, stores them as JSON, and pre-generates static HTML pages that can be dropped into any site as a responsive, auto-sizing iframe.

Reviews are fetched via the Apify API using the tokens stored in GitHub Actions secrets. Rendered HTML is published to a CDN, and a lightweight embed script pairs an <iframe> on the host site with its iframe-resizer child running inside the iframe so the frame grows to match its content.

An array of businesses to fetch, each with the platform identifiers it has and per-platform fetch state. A minimal entry looks like:

{
  "google_business_id": "ChIJ...",
  "facebook_page_url": "https://www.facebook.com/Example",
  "trustpilot_url": "https://uk.trustpilot.com/review/example.com",
  "slug": "example",
  "number_of_reviews": -1,
  "minimum_star_rating": 0,
  "fetch_frequency_days": 7
}

A business can include any subset of the three platform fields. Each platform that has been fetched records its own last_fetched_<source> timestamp so the three sources can run on independent schedules.

Per-platform fetchers. Each one loads config.json, filters to businesses that have the relevant platform field, checks whether fetch_frequency_days has elapsed since the last fetch for that source, and then calls the appropriate Apify actor. Results are normalised to a shared review shape and written to data/<slug>/<author-slug>-<YYYY-MM-DD>.json.

Each script accepts an optional slug argument to fetch just that business:

bun src/fetch-google-reviews.js my-business-slug

Shared helpers: config loading, Apify HTTP calls (with a curl fallback for flaky DNS), reviewer avatar download + resize to .webp thumbnails via sharp, filename formatting, and the review fetcher factory that the three platform scripts are built on top of.

Reads the JSON files for each business, deduplicates reviews that appear on more than one platform (priority: Trustpilot > Google > Facebook), sorts the remaining reviews newest-first, and renders data/<slug>/index.html and data/<slug>/code.txt (the <iframe> embed snippet to paste into a host page).

Can be called with a single slug to render just that business or with no argument to render all of them.

The iframe template. It includes inline CSS, inline SVG, and the bundled masonry.js and iframe-resizer-child.js so that everything a visitor needs arrives in a single HTML request. Review cards flow in a masonry column layout that reshapes to fit the iframe width.

The embed side of the integration. src/build.js uses Bun's bundler to produce three minified browser scripts in dist/:

• reviews-embed.js - loaded by the host site; wires iframe-resizer's parent side up to any iframe on the page.
• iframe-resizer-child.js - loaded inside the iframe so it can report its content size back to the parent.
• masonry.js - client-side greedy shortest-column masonry layout that runs inside the iframe.

Run the bundle locally with:

bun run build

Generated output, committed back to the repo by the GitHub Action. Per business:

• data/<slug>/*.json - one file per review.
• data/<slug>/index.html - the rendered iframe contents.
• data/<slug>/code.txt - the embed snippet for that business.

Downloaded reviewer avatars, resized to 48x48 (<userId>.webp) and 96x96 (<userId>@2x.webp) WebP images. Served alongside the rendered iframes from the CDN.

The scheduled GitHub Action. Runs daily (and on push) to:

1. Install dependencies and build the embed bundles.
2. Run each platform fetcher (Google, Facebook, Trustpilot) for every business that is due.
3. Render the iframes.
4. Commit any new reviews/renders back to the repo.
5. Sync data/ and the embed script to Bunny CDN.

The workflow can also be dispatched manually with an optional business slug and source (google, facebook, trustpilot, or all) to refresh a single business or a single platform.

Many customers leave the same review on more than one platform. At render time, reviews are deduplicated across sources by normalised author name and content. When the same review is present on multiple platforms, the highest-priority copy is kept:

trustpilot > google > facebook

The source-specific JSON files are left untouched on disk; deduplication only affects what is rendered into the iframe.

bun install                       # install dependencies
bun run build                     # bundle embed scripts into dist/
bun run fetch                     # fetch Google reviews (needs APIFY_API_TOKEN)
bun run fetch:facebook            # fetch Facebook reviews
bun src/fetch-trustpilot-reviews.js
bun run render                    # regenerate data/<slug>/index.html
bun run lint                      # biome check
bun run test:e2e                  # playwright end-to-end tests

An APIFY_API_TOKEN in a .env file at the repo root is required to fetch from any of the three platforms.