skylite

Skylite — product concept & branding

What Skylite is (one line): a read-only, non-algorithmic window into Bluesky/ATProto built for kids — “Skylite” = *Bluesky* + lite.” The pitch: strip out posting, replying, and DMs and you remove the large majority of open-social child-safety risk while still offering a curated window into global curiosity.

Provenance: distilled from a Gemini design dialogue pasted on 2026-07-10 — raw at seeds/transcripts/raw/2026-07-10-gemini-skylite-concept-and-logos.md. This doc holds the product/experience + branding thinking; the identity/ATProto/PWA-hardening engineering lives in IDEAS.md (a separate same-day session). Nothing here is verified engineering — it is a captured concept. Claims sourced only from the model’s framing are flagged model-claim and should not be treated as measured.

---

1. Positioning

• Read-only by construction. No posting, no replying, no quote-posts, no DMs — purely passive consumption. The safety argument is that removing the interaction surface removes most of the abuse surface (harassment, grooming via DMs, algorithmic rabbit-holes). The “~95% of risk eliminated” figure is model-claim, not a measured number — treat it as directional framing, not data.
• A lens, not a new network. Skylite is a specialized client over the existing ATProto universe, not a from-scratch social network. This mirrors the IDEAS.md architecture (client-side merge of an inclusion list) rather than contradicting it — see the axis note in §2.
• Not built to be addictive. No follower counts, no engagement metrics, no “rage-bait” loops. Bright, high-contrast, large-text, highly visual “Lite” UI.

2. The reading experience — two framings on the same axis

The Gemini dialogue framed discovery as curated “Sky-Channels” (Science, Art, Storytime) built from custom ATProto feeds curated by educators/creators (e.g. #SpaceStuff, #CuteAnimals, #KidArt, #DIYCrafts). IDEAS.md §2 lands on a lighter, serverless framing: an inclusion list merged client-side with getAuthorFeed, newest-first, no server.

• These are the same design axis at two power levels, already named in IDEAS.md: a server-side custom feed generator (more power, a service to operate) vs. a client-side merge (lighter, nothing running). “Sky-Channels” = the curated, server-side end; the inclusion list = the serverless end. Not a contradiction — a cost/capability trade to decide, not re-litigate.
• Shared ceiling either way: the child only ever sees what a guardian included / which channels are on. That is the point and the limit.

3. Feature set (concept)

Area	Concept
Interaction	“Save to Scrapbook” — private, local bookmarking in place of like/repost. Kid can clip a post and add private local notes (“want to draw this later”).
Discovery	Curated Sky-Channels (Science / Art / Storytime), toggled per-child.
Moderation	“Sky-Shield” — an automated layer on top of ATProto labels: image-recognition filtering + text filters that blur profanity/politics/sensitive-news even from an approved creator.
Co-viewing	Cast the feed to a TV/tablet — “scrolling” becomes a shared family activity, like reading a digital magazine together.
Guardian oversight	Dashboard to toggle channels on/off and view watch history.

Cross-check with IDEAS.md: the Scrapbook, Sky-Channels, Sky-Shield, co-viewing, and the toggle/watch-history dashboard are new here — IDEAS.md does not cover them. Where they meet the technical doc: the dashboard’s per-child toggle is the product surface of the IDEAS.md pause screen / kill switch (client flag), so those should be designed as one control, not two. The Scrapbook is local-only, which fits the serverless stance in IDEAS.md §2.

Tension to hold (not resolve here): “Sky-Shield” AI image/text filtering implies running a service on incoming content — directly in tension with the “no server, client-side, nothing running” spine of IDEAS.md. Any real image-recognition moderation likely reintroduces the confidential-backend question that IDEAS.md §5 already flags for session length. Flag, don’t decide.

4. Branding & naming decision (see NAMING note in this doc’s §6)

• Name: Skylite — a play on Bluesky integration + lite.
• Winning visual (converged over ~4 logo rounds): a rectangular skylight set into a ceiling, viewed from below (lying/standing under it), looking up through the glass into a deep-indigo night sky — crescent moon, stars, one soft cloud. The “Cosmic Window” cosmic elements carried through every round.
• Tagline: “A Window to the Stars.”
• Palette: bright + dark blue, yellow, white — safe, friendly, kid-appealing.
• Secondary theme explored: day/night (sun-half / moon-half) — parked in favor of the night-sky skylight.
• Perspective the user actually wanted (worth recording so it isn’t re-derived): looking up at a skylight on the ceiling, with a short end of the rectangle nearest the viewer (parallel to your feet / a board laid across your toes), the long sides receding to a vanishing point — a long, narrow rectangle stretching away overhead. The final dialogue turn was the user disambiguating this geometry, not a new concept.

5. Open product questions (carried forward)

1. Target age band — unresolved, the model asked and it was not answered: 6–9 (heavy curation) vs. 10–12 (stepping-stone to real social). This shifts curation depth, UI, and the graduation timeline in IDEAS.md §1.
2. The Boredom Factor. Without the slot-machine of interaction, does a read-only app hold a kid, or get abandoned for YouTube? The core product risk.
3. Monetization. Model floated ~$2/month parental subscription, not ads (ads on a kids’ app is a non-starter; AI safety filtering costs money). model-claim on cost; a real number needs the moderation architecture pinned (see §3 tension).
4. Does “Sky-Shield” survive the serverless spine? (§3) — the make-or-break collision between the moderation promise and the no-server identity/PWA design.

6. Naming decision (durable)

Decided in-dialogue (2026-07-10):

• Product name Skylite (Bluesky + lite).
• Identity/visual: skylight → night sky, tagline “A Window to the Stars.”

These are branding decisions, not verified engineering. Recorded here so a later session treats the name/visual/tagline as settled and doesn’t re-explore the logo space. The geometry the user wanted is in §4.

This site is open source. Improve this page.