nuxt.fyi

Spots Nuxt sites being shared on Bluesky and posts screenshots of them.

👉  Check it out

The aim of nuxt.fyi is to celebrate sites built with Nuxt in the wild.

It watches the Bluesky Jetstream firehose, pulls every link out of every post, and for each new domain asks: is this built with Nuxt? When the answer is yes, it posts a screenshot on Bluesky.

The public dashboard shows everything that’s been detected, broken down by version and recency.

Features

• Detection signals across HTML and a Playwright runtime probe (<div id="__nuxt">, __NUXT_DATA__, window.__NUXT__, meta[name=generator], /_nuxt/ assets)
• Screenshots taken in a persistent Chromium context, with layered cookie-banner suppression (GTM consent cookie, CSS overlay, and a pinned I-Still-Dont-Care-About-Cookies extension)
• Discord + Bluesky notifications with rate-limit-aware posting
• @-mention triggered replies: tag the bot on Bluesky with a URL and it replies with the scan result + screenshot, scanning on the spot if it’s a domain we haven’t seen yet
• Nuxt 4 dashboard reading the same SQLite database read-only
• Runs on a single Fly.io machine; daemon and dashboard share one volume
• Node 24+ only: native TS execution, node:sqlite, global fetch and WebSocket

Try it out locally

You will need a Discord webhook URL and, optionally, a Bluesky account with an app password for posting.

cp .env.example .env
# edit .env, set DISCORD_WEBHOOK_URL (and BLUESKY_* if you want)

corepack enable
pnpm install
pnpm install-extension
pnpm --filter @nuxt-fyi/scanner exec playwright install chromium

# scanner (terminal 1: screenshot service on :3001)
pnpm dev:scanner

# daemon (terminal 2)
pnpm dev

# dashboard (terminal 3)
pnpm dev:dashboard

Set VERBOSE=1 to log every post the daemon sees.

Manual submissions

The dashboard’s home page has a small form for queueing a domain by hand. Under the
hood:

• The dashboard’s POST /api/submit rate-limits per IP (default 5 / minute, tunable via
  SUBMIT_RATE_LIMIT and SUBMIT_RATE_WINDOW_MS).
• It then posts the URL to the daemon’s loopback-only submit server (DAEMON_SUBMIT_PORT,
  default 3010) using the shared DAEMON_SUBMIT_TOKEN bearer secret.
• The daemon canonicalises the domain, checks the global skip list, and either enqueues a
  detection job (same path the Jetstream consumer uses) or short-circuits with the
  existing result if the domain has been scanned within RESCAN_AFTER_MS.

Set DAEMON_SUBMIT_TOKEN as a Fly secret on nuxt-fyi to enable it in production; leave
it empty to disable submissions entirely (the daemon refuses every request, the dashboard
returns 503).

NSFW classification

Screenshots are classified with nsfwjs on the
scanner machine at capture time. Each row gets one of three labels:

• safe (default): renders normally everywhere.
• suggestive: renders normally on the dashboard; Bluesky posts get a sexual self-label
  so users can opt to blur via their account preferences.
• nsfw: dashboard blurs the image with a click-to-reveal overlay; Bluesky posts get a
  porn self-label; Discord posts attach the image as a SPOILER_* file rather than
  rendering inline in the embed.

Thresholds are tunable via NSFW_PORN_THRESHOLD (default 0.5) and NSFW_SEXY_THRESHOLD
(default 0.6) on the scanner. After tweaking, re-run the backfill with --reclassify to
relabel historical rows.

Architecture

Three processes split across two Fly apps:

• nuxt-fyi (this fly.toml): daemon (Jetstream consumer + detection) and dashboard
  (Nuxt 4) in one container, sharing the SQLite volume at /data.
• nuxt-fyi-scanner (fly.scanner.toml): Nitro v3 service that owns Playwright + the
  screenshot half of the ImageKit upload. No public ingress; daemon reaches it on
  nuxt-fyi-scanner.internal:3000 via Fly’s 6PN private networking. Stays always-on at
  shared-cpu-2x because .internal doesn’t wake stopped machines.
• ImageKit: external CDN for screenshots + og:images.

The daemon talks to the scanner via an authenticated HTTP call (SCANNER_TOKEN shared
secret). Scanner outages degrade quality (no screenshot) but don’t break the pipeline:
the og:image is still uploaded and the hit is still recorded + posted.

Image hosting

Screenshots and og:images are uploaded to ImageKit at scan time
so the dashboard can render them through @nuxt/image with on-the-fly resizing. The
dashboard renders ImageKit URLs only; for the rare row where the daemon recorded an
og:image origin but the upload didn’t land, a plain <img> falls back to the upstream
URL.

Backfill scripts

Two idempotent scripts on the daemon side, safe to re-run:

# Rescan Nuxt-confirmed rows missing an ImageKit screenshot, or with an og:image URL
# recorded but no ImageKit copy. Uses the live scanner; ignores RESCAN_AFTER_MS.
pnpm backfill-images --concurrency=2

# NSFW-classify every row that has an image but no nsfw_label. Calls the scanner's
# /classify endpoint. Add --reclassify after threshold tweaks to relabel all rows.
pnpm backfill-nsfw --concurrency=2

Both support --limit=N and --dry-run.

Admin CLI

On the running Fly machine (fly ssh console -C bash), you can re-scan one or more
domains against the live database, or check what we currently know about one:

cd /app
node src/cli/rescan.ts example.com another.com

# refresh only the screenshot/og:image on an existing Nuxt hit
node src/cli/rescan.ts --screenshot-only example.com

# suppress Discord + Bluesky posts even when the domain is newly detected as Nuxt
node src/cli/rescan.ts --no-notify example.com

# read-only: dump the stored scan, NSFW classification, activity, and notifications.
# Pass --json for machine-readable output.
node src/cli/status.ts example.com

Credits

Detection heuristics and the GTM consent-cookie trick are lifted from nuxtlabs/vue-telescope-analyzer ❤️

License

Made with ❤️

Published under MIT License.