J
Jordan Lee
Photo · skate park, Saturday session
Finally landed the kickflip I've been working on since August. Only took about 400 tries and one very bruised elbow.
Parents sign up first. Teens can't get in any other way.
There is no teen-only sign-up. A parent verifies their own identity, then creates and approves the teen's account. Until all four steps are done, the teen account stays inactive. Click through the flow — it's fully simulated.
- 1Parent account
- 2ID check
- 3Live selfie match
- 4Create teen account
Create your parent account
This is the account that approves contacts, receives safety alerts, and can pause the teen's account at any time.
Fictional example details are pre-filled so you can click through. Nothing is sent or stored.
Verify your government ID
In the real product, a verification provider checks the document's security features. Here, the check is simulated with a placeholder card — no document is uploaded.
Placeholder — sample driver's licence graphic. Real ID images would go only to the verification provider, never to this app's own database.
Ready to run the simulated check.
Live selfie — does the face match the ID?
The provider compares a live selfie against the ID photo and checks liveness (a real person, not a photo of a photo). Simulated below — your camera is never opened.
Waiting to run the simulated match.
Face-match confidence
What the app keeps from this step: a single "parent verified ✓" token and the provider's check ID. The selfie and ID image stay with the provider and are deleted on their retention schedule — the app never stores face data.
Create & approve the teen account
The teen account is created by you, linked to you, and only becomes active when you approve it here.
Ages 13–17 only. Under 13 isn't allowed (COPPA); at 18 the account converts or closes, with the parent notified.
Message scanning — please read. This is also stated in the Terms of Use, word for word:
All conversations are scanned by AI by default for patterns linked to grooming and exploitation. No human reads your teen's messages — no staff, no moderators; the scanning is done by software only.
If the AI detects a threat, the conversation is paused and you, the parent, are alerted. You can then review the flagged messages in your parent dashboard and decide whether the contents are appropriate for your child. If they are not, you can call the police — and the app gives you a direct link to the NCMEC CyberTipline with every high-risk alert.
You can only ever access messages that get flagged. Unflagged conversations are never shown to you, or to anyone at the app.
✓
Account active
Maya's account (@maya.makes) is live and linked to Dana Whitfield (parent, identity verified). Maya can now sign in — try the teen app and the parent dashboard.
The teen app, as Maya sees it
A normal-feeling social app — feed, messages, videos, profile. Safety runs underneath: contacts need parent approval, messages are scanned quietly (disclosed at sign-up), and no video posts until its pre-post scan passes. Open Messages to run the scanner demo — try the conversation with Lucía in Spain to see the translation bot — open Videos to run the pre-post scan, or open Learn for the math tutor, live lessons, the whiteboard, the grammar cheat sheet, and the writing scanner.
M
Maya
@maya.makes · 15
M
P
Priya Nair
Book club pick for November is out and I have OPINIONS. No spoilers but chapter 19?? Chapter 19.
S
Sam Ortiz
Photo · school talent show rehearsal
Our band got the closing slot at the talent show. Rehearsal photo dump — yes the drummer is asleep in one of them.
T
Tessa Brooks
Attempt #3 at my grandma's lemon bars. The powdered sugar ratio is finally right. Recipe trade in the comments?
A
Alex R.
New contact · messages are safety-scanned (see disclosure)
This conversation has been held by the safety scanner. No new messages can be sent. Maya's parent has been alerted and can review these flagged messages in the parent dashboard.
Safety scanner demo
Run a sample conversation through the scanner, or type a message and score it. The scanner looks for patterns building up — not single bad words.
Conversation risk score: 0/100
Safe
0 · Safe40 · Flagged for parent review70 · Held + parent alert
Try: "don't tell your parents, I'll send you a gift card if you send me a pic" — then check the Parent dashboard tab to see the alert arrive.
Upload a video
Nothing posts until the pre-post scan finishes. The scan checks your video against the federal baseline and the rules of the state your account is registered in — pick a different state to see how the rules change.
- Reading the video — frames, audio track, and on-screen text
- Checking against the federal baseline (applies in every state)
- Applying your state's rules overlay
- Decision — approved, held for your parent, or blocked
Simulated scan — no file is uploaded and no video leaves this page. The state rules shown are design examples; the production rulebook is written per state from the law catalog (see Rollout & law catalog).
Fresh from teen creators
Skate session
3:42Book talk
8:15Live set
5:03Baking
6:27Workshop tour
4:18Ceramics
2:36Your earnings — this month so far
Videos earn from ads shown alongside them, split by John's formula: creator earning = ad revenue − cost to run − 10% markup. Full calculator on the Videos tab.
Ad revenue
$128.40
$128.40
Cost to run
− $38.15
− $38.15
Markup (10%)
− $12.84
− $12.84
Your earning
$77.41
$77.41
Payouts go to the parent-managed account linked to your account — minors can't hold ad accounts directly, so earnings land with your parent and they release them to you. You can always see the full ledger here.
M
Maya Whitfield
@maya.makes · 15 · joined October 2026
Parent-linked account Contacts: parent-approved only
Sophomore. Ceramics, track, and filming my friends' bands. If I don't know you from school, my mom has to approve you before we can talk — that's the rule here and honestly it's fine.
128
Posts
86
Approved friends
94
Following
Interests
CeramicsTrack & fieldIndie bandsPhotographyBaking
Privacy defaults for every teen account: profile visible to approved contacts only, no public follower counts from strangers, no location in posts, and no discoverability by phone number or email for anyone outside approved contacts.
Learning tools
Help with your own work — a step-by-step math tutor, lessons with your linked parent/tutor, a shared whiteboard, a grammar cheat sheet, and a scanner that checks your writing. The writing scanner is a learning helper for your own text — it is separate from the safety scanner, which looks for grooming patterns in conversations.
Live lesson
No lesson running
No live lesson yet. Lessons appear here when your linked parent or tutor starts one from their account — lessons only happen between accounts that are already linked, never with strangers.
Try it: start a lesson from the Parent dashboard tab, or use the demo button in the Learning tab.
Math tutor AI
Works problems with you step by step — a hint first, you answer, then the next step. It never just dumps the final answer.
Virtual whiteboard
A shared board inside a lesson — you and your tutor both draw and write on it. It actually draws: try it in the Learning tab.
Grammar cheat sheet
Quick-glance cards for the mix-ups everyone makes — their/there/they're, your/you're, then/than, and more.
Writing scanner
Paste your own writing and it catches common misspellings and wrong-word use — and explains why each fix is right.
Parent dashboard — Dana Whitfield
Everything a parent can see and do: the linked teen account, safety alerts (including ones the scanner demo generates live), contact approvals, and account controls.
Linked teen account
M
Maya Whitfield
Pause account
Posting and messaging stop immediately. Maya can still open the app and see this explained.
New contacts need my approval
On by default and can't be turned off by the teen.
Weekly activity summary
New friends approved, time on app, and any scanner flags — every Sunday. Not a feed of Maya's posts or routine messages.
Lessons — teach Maya from your account
Start a live lesson with your linked teen. The lesson runs from your account straight into Maya's — the math tutor and the shared whiteboard open for both of you. Lessons only work between accounts that are already linked (your verified parent account ↔ Maya's teen account). No one outside the link can start, join, or watch a lesson.
No lesson running
When you start, a lesson card appears in Maya's app (Teen app → Learn) and the lesson room — tutor + whiteboard — opens in the Learning tab.
How message scanning works — as stated in the Terms of Use
Message scanning disclosure (shown to parent and teen at sign-up, in plain language):
"All conversations are scanned by AI by default, looking for patterns linked to grooming and exploitation — things like requests for secrecy, gifts or money offers, asking for photos, pressure to move to another app, or plans to meet or travel. No human eyes ever read these messages — not app staff, not moderators. The scanning is done by software only."
"If the AI detects a threat, the conversation is paused and you, the parent, are alerted. You can review the flagged messages right here in your dashboard and decide whether the contents are appropriate for your child. If they are not, you can call the police — and every high-risk alert includes a direct link to the NCMEC CyberTipline."
Flagged messages only: you can access a conversation only when it has been flagged. Unflagged conversations are never shown to you — or to anyone at the app. Because messages are AI-scanned, they are not end-to-end encrypted; that trade-off is stated up front, in the Terms of Use and at sign-up, not buried.
Safety alerts
ReviewContact request held for your approvalTue · 4:12 PM
An account claiming to be a classmate ("alex.r_918") asked to message Maya. It isn't matched to any school or mutual contact, so it was held for you instead of going through.
Run the grooming-pattern scenario in the Teen app → Messages tab and a new high-risk alert will appear here.
If a flagged conversation is not appropriate for your child: call the police, or report it to the NCMEC CyberTipline. This link is included with every high-risk alert. You can only ever see messages that were flagged — nothing else in Maya's account is visible here.
Contact requests
A
Alex R. @alex.r_918
N
Nico Alvarez @nico.a
Maya's videos — review queue
Every video is scanned before it can post — federal baseline plus your state's rules. Approved videos post on their own. Held and blocked videos come to you here, under the same rule as messages: you only ever see videos that were flagged. Maya's approved videos are never shown in this dashboard.
Held for your review"Backyard ramp jump — second angle"Today · 3:48 PM
The pre-post scan finished, but the state-rules overlay flagged this one: Oklahoma overlay — dangerous stunt rule (a ramp jump filmed without a helmet). It has not posted. Watch it and decide: approve the post, or keep it blocked.
Waiting for your decision. Maya sees "held for parent review" — not your reasoning.
Blocked — never posted"Bottle-rocket launch, slow-mo"Tue · 6:02 PM
Blocked by the scan before posting: fireworks content under the Oklahoma overlay (and fire-starting content is blocked under the federal baseline too). Blocked videos can't be approved from here — the decision is final, and you're notified so you can talk to Maya about it.
Video earnings
Maya's videos earned $77.41 this month so far (ad revenue $128.40 − cost to run $38.15 − 10% markup $12.84). Payouts land in your parent-managed account — minors can't hold ad accounts directly — and you release them to Maya. She sees the same ledger in her app.
How this scales — and what's real vs. simulated
The app itself is the easy part. The honest costs of a teen-only network are verification, secure preservation of flagged messages as evidence, and legal duties — those don't scale for free.
The app layer scales itself
Feed, profiles, posts, follows, and messages run on managed services — Cloudflare for the site and media delivery, a Supabase-class backend for accounts and data. Capacity grows automatically; there are no servers to babysit, and you start on free/cheap tiers and pay more only as usage grows.
Safety scanning is software
Every message is scored as it's sent — pattern categories (secrecy, gifts, photo requests, app-switching, meet/travel plans), weighted and combined over a whole conversation. Scoring is cheap compute and scales with traffic. A high score pauses the conversation and alerts the parent — the parent does the reviewing, so review capacity grows with families, not with headcount.
Trust is the bottleneck
Each sign-up costs a real ID check, and each flagged conversation has to be stored securely and preserved intact so the parent can review it — and hand it to police or the CyberTipline as evidence if they choose. Those two line items — not hosting — decide whether the service can grow safely.
The real costs (per unit, not per server)
- ID + face verification: roughly $1–$2+ per parent sign-up at typical provider pricing. Free for families, never free for the operator — at 10,000 sign-ups that's $10–20k before anyone posts anything.
- Flagged-message handling: flagged conversations must be stored securely and preserved untouched, so the parent can review them and pass them to police or the CyberTipline as evidence. Parents also need support when an alert turns out to be a false alarm.
- Video running costs: every video costs real money to store, stream, and scan. That per-video cost is metered precisely because it's a line item in the creator earnings formula — a creator's "cost to run" is their videos' actual hosting, bandwidth, and scan compute for the month, not a flat fee.
- Legal groundwork: ages 13–17 only (under-13 triggers COPPA), state teen social-media parental-consent laws vary and are shifting, and minors' data needs strict retention limits. Places open one at a time: each state and country gets a law-catalog entry that is reviewed and configured into the app before it becomes available — see Rollout & law catalog.
Known limits, stated plainly
- Not end-to-end encrypted. Scanning and encryption can't both be absolute; the product chooses scanning and says so at sign-up.
- False positives happen. Jokes and song lyrics can trip patterns — that's why a flag only pauses the conversation and alerts the parent, who knows the child and makes the call. Nothing triggers an automatic accusation or ban.
- Determined bad actors adapt. Pattern scoring raises the bar and catches the common scripts; it is not a guarantee, and parent controls are a second layer, not decoration.
Goes live country by country, state by state
EyePost doesn't launch everywhere at once. Each place — a US state, then a country — goes through the same simple pipeline: catalog its laws for review, configure the app to be legal there, then make it available. No lawyer gate. The catalog is the review. A place stays closed until its catalog is complete and its settings match it.
1
Catalog the laws
For that state or country, list the actual rules that touch this app: age floor, parental consent, ID / biometric rules, data rules, reporting duty, and the local tip line. Each rule goes in the catalog with its source law named, in plain language, ready for anyone to review.
2
Review the catalog
The catalog entry is checked line by line — is anything missing, out of date, or marked "not settled yet"? Unsettled items keep the place closed. No outside sign-off is required; the completed, reviewed catalog is the gate.
3
Make the app legal there
Turn the catalog into settings for that place only: its age floor, its consent wording and who can consent, its state / country overlay for video scanning, its preservation timer, and its tip line in every alert. Nothing goes live on default settings.
4
Make it available
Flip that one place to Available. Sign-ups open for accounts registered there — and nowhere else new. Every live place is re-cataloged when its law changes, and can be paused with the per-place kill switch.
Rollout board — pick a place to review its law catalog
This is the working board for the rollout. United States first, state by state — then country by country. Status means exactly what it says: Available = catalog reviewed + app configured + open for sign-ups there. In review = catalog built, being checked. Cataloging = laws being collected, still closed. Tap a place to read its catalog entry.
United States — state by state
Countries — one at a time, after the US
Prototype board with design-example entries. In production this board is driven by the real catalog: one entry per place, each law named with its source, reviewed in the open, and wired straight to that place's app settings.
What every catalog entry must contain before a place can open
A place moves to Available only when every line of its catalog entry is filled in and reviewed:
- Age floor and digital-consent age, with the law that sets them
- Parental-consent mechanics — who counts as the consenting adult, what form consent takes, and what proof is kept
- ID / face-match rules for that place, plus an alternative non-ID check where the law requires one
- The local tip line for every parent alert, and any mandatory reporting duty, with its preservation period
- Whether AI safety scanning is allowed there, and where the data may be stored / transferred
- Terms, consent screens, and parent alerts ready in the place's language(s)
- Status flipped to Available — by the catalog review, not by a lawyer. If a law is unsettled, the entry says so and the place stays closed.
Every live place is re-checked quarterly, and immediately when its law changes — this area of law is moving monthly.
Not served — geo-blocked
- North Korea. Comprehensive sanctions leave no lawful way for a US-based service to operate there.
- China. Real-name registration for minors, plus state data-access and content-control duties, clash head-on with this design: pseudonymous teens, parents seeing flagged messages only, no state role in anyone's chats.
- Russia. Law requires Russian users' data to be stored on servers in Russia first, with a local office answerable to the regulator — the opposite of this service's data model. Sanctions compound it.
All three are blocked at the network level and named in the Terms of Use.
Every alert carries the local tip line
The CyberTipline is a US service. A parent in another country gets that country's reporting route in the alert itself:
- US: NCMEC CyberTipline · Canada: Cybertip.ca · UK: CEOP / IWF
- EU: the member state's INHOPE hotline · Australia: eSafety / ACCCE ReportCyber
- New Zealand: Netsafe · Brazil: SaferNet Brasil · India: the national cyber-crime reporting portal
- Japan, South Korea, South Africa, Mexico, UAE: several countries have no single tip-line equivalent — the catalog records the correct police / regulator route before the place opens, so the alert still gives the parent a real, local next step.
Watch list — not settled law
- Mexico: new data law plus a pending written-consent reform — verify against the official text at launch.
- Canada: federal online-harms and under-16 bills have been tabled but are not confirmed as law.
- New Zealand: the under-16 ban is a bill, not law — its status is re-checked in the catalog before opening.
- Japan: under-16 consent and biometric rules commence in stages through 2028.
- EU / France: an EU "Kids Act" and a replacement French text are proposals, not law.
These countries are marked "watch" in the catalog — the prototype doesn't present unsettled law as settled.
One narrow exception to "no human eyes" — a US legal duty
US federal law (18 U.S.C. § 2258A, expanded by the REPORT Act) requires providers to report apparent child sexual abuse material and online enticement to the NCMEC CyberTipline once they have actual knowledge of it — and to preserve the reported content for one year. So when the AI flags apparent CSAM or enticement, a small, access-controlled safety-escalation path files that report: flagged content only, every access logged. It is not a staff reading queue, nobody browses conversations through it, and it exists only where the law requires it. Parents are still alerted, and parents still do the reviewing for everything else.
Design commitments the research forces
- Age floors are settings, not one global 13–17. Each country's floor and consent age lives in a configuration table (Australia 16, India/South Africa consent to 18, EU per member state) and can change without a code release.
- Biometrics stay vendor-side. The verification provider performs the face match; the app stores only a pass/fail token — never a face template. And teens are never face-scanned: verification is of the parent only.
- Per-country consent settings. The consenting adult's role, the exact wording, and the proof kept differ by country ("competent person" in South Africa, "legal representative" in South Korea, verifiable consent in India) — consent is stored per jurisdiction.
- Per-country preservation timers. Flagged content is kept as long as that country's law requires (one year for US-reported content) — not on one global delete timer.
- No ads, no profiling of teens — permanently. The EU, India, Brazil, and the UAE all ban profiling-based advertising to minors; staying ad-free removes a whole compliance layer in every market.
- An 18th-birthday re-consent flow. Where a parent's consent carried the account, the now-adult user re-consents in their own right at 18 and the parent link converts or closes.
- A per-country kill switch. Any country can be paused — or its age floor raised — in hours, so a new law (the New Zealand scenario) is complied with immediately, not next release.
Catalog snapshot dated 23 October 2026 covering 14 markets plus the three blocked countries. Each entry names its source laws for review; a place opens only when its entry is complete, reviewed, and configured into the app — the catalog is the gate.
Every language where it could be used — with a translation bot for cross-country chats
The goal: a teen can use EyePost fully in their own language, and talk to a teen in another country without either of them speaking a foreign language. This page is the design for that — plus one working demo conversation. What is built here, and what is deliberately deferred, is stated plainly at the bottom.
Languages launch with the country gate — never half-translated
Language is part of the same place-by-place gate as the law catalog. A country does not go live until all of this works in its main language(s):
- The whole interface — feed, messages, profile, settings, and the language picker itself.
- The Terms of Use and the sign-up disclosures — including the message-scanning disclosure a parent accepts.
- The parent dashboard and every safety alert — a parent must be able to read an alert in their own language, first time, under stress.
- Legal text is professionally translated — Terms, consent screens, and disclosures are translated by professional translators and checked against the place's catalog entry. Never machine-only.
The first language set follows the researched markets: English, Spanish, Portuguese, French, German, Italian, Dutch, Polish and other EU languages per member state, Korean, Japanese, Hindi and major Indian languages, Arabic, and Afrikaans and other South African languages as feasible. "Every language where it could be used" is the end state; rollout is per-country.
The translation bot — how a cross-country chat works
- Each teen writes in their own language. Maya writes English; Lucía in Madrid writes Spanish. Neither needs the other's language.
- Every message is auto-translated for the reader and labelled plainly: "Translated from Spanish." No silent rewriting — the reader always knows they're reading a translation.
- One tap shows the original. "See original" reveals the message exactly as written, so nuance, jokes, and wording can be checked.
- A bot line at the top of the thread explains it before the conversation starts — who writes in what, and that translation is on.
Opens the Teen app → Messages. Her Spanish messages appear translated into English, each with a "See original" toggle.
The scanner reads the original — translation never touches safety
- Scanning happens on the message in its original language, before any translation. The bot can never launder a grooming message into a "clean" translation — the scanner has already read what was actually written.
- Each language gets its own pattern set. Grooming language is slang-heavy and culture-specific: secrecy, gifts, photo requests, app-switching, meeting and travel talk each need patterns built and tested in that language. An English-only scanner reading translations would miss coded, local phrasing.
- Mixed-language messages are scanned in each language present.
- Translations are for reading — never the evidence. Flagged messages are preserved in their original wording, exactly as sent.
What a parent sees when a translated chat is flagged
The flagged message is shown in the parent's language, with the original directly underneath — so nothing is lost between the alert and the evidence, and a police or tip-line report carries the original wording. Design example (fictional):
Shown to the parent (translated from Spanish): "Don't tell your mum about this chat, okay? It stays between us."
Original — Spanish, preserved exactly as written"No le digas a tu madre lo de este chat, ¿vale? Esto queda entre nosotros."
The flagged-only rule doesn't change with language: parents still see only flagged messages, and the local tip line for the account's country comes with the alert.
Right-to-left languages
Arabic launches with a full right-to-left layout — the whole interface mirrors: navigation, message bubbles, and controls all read right-to-left, not just translated text dropped into a left-to-right page. RTL support is a launch-gate item for Arabic-speaking markets, tested with native readers before the gate passes.
Deferred on purpose — not built in this prototype
Following the "if it takes a long time, skip it for now" rule: the full production translation engine, the per-language scanner pattern libraries, and the professional legal translations are deferred launch-gate work for each country — scheduled alongside that country's compliance pass. They are not built, and this prototype doesn't claim they are. What this prototype shows is the design on this page, the language pickers on the teen app and parent dashboard (preview labels only), and the one demo conversation with Lucía.
Videos — a YouTube-style service for teens, scanned before anything posts
Teens entertain each other with videos: skits, music, crafts, games, sports. The difference from YouTube is the door every video walks through — a pre-post scan against the law of the poster's own state — and who gets paid what, by a formula stated in plain words below.
Nothing publishes until the scan finishes
Uploading is not posting. Every video goes through the same pipeline, in this order, and the Post button doesn't exist until the pipeline says so:
- 1 · Read the whole video. Frames are sampled, the audio track is transcribed, and on-screen text is read — a violation spoken or written counts the same as one shown.
- 2 · Federal baseline — every state, no exceptions. Sexual content involving a minor (always blocked, preserved, and reported to the NCMEC CyberTipline under the §2258A duty — never a parent-review option), real violence or serious injury, instructions for weapons or explosives, drug manufacture or sale, and content promoting self-harm.
- 3 · The poster's state overlay. On top of the federal floor, the scan applies the rules of the state the poster's account is registered in — stunt and challenge rules, fireworks, wildlife, weapons-handling, and product rules differ by state. The upload box in the teen app has a state picker: change the state and the rules list changes.
- 4 · One of three outcomes — approved, held, or blocked (below). There is no "post now, scan later," and no way for a teen to skip or appeal around the scan itself.
Opens Teen app → Videos: run the simulated scan, change the state, and see the earnings preview.
Approved → it posts
The scan found nothing against the federal baseline or the state overlay. The video goes live for approved contacts, with a "pre-post scan passed" marker. Most videos land here.
Held → parent decides
Something tripped a state rule or a judgement call (a stunt without a helmet, a replica prop). The video does not post — it goes to the parent dashboard, where the parent watches it and approves or keeps it blocked. Same flagged-only rule as messages: parents see held videos, never a teen's approved ones.
Blocked → it never posts
A clear violation of the federal baseline or a hard state rule. The video is stopped permanently, the parent is notified, and there's no approve button — the parent's role is to know, and to talk to their teen. Apparent CSAM is preserved and reported, per federal law.
Creator earnings — John's formula
Creator earning = Ad revenue − Cost to run − 10% markup
In plain words: from the ad money a teen's videos bring in, first subtract what those videos cost to run — hosting, streaming bandwidth, and the pre-post scan. Then subtract the 10% markup — the platform's cut, calculated as 10% of the ad revenue. Everything left is the teen creator's earning.
This is the formula as designed from John's wording — if the split doesn't match what he meant, this wording is the thing to correct, and the calculator below follows it exactly.
What the ads shown with the videos brought in.
Hosting, streaming, and scanning for those videos.
Ad revenue$100.00
Cost to run− $30.00
Markup (10% of ad revenue)− $10.00
Creator earning$60.00
Worked example: $100.00 − $30.00 − $10.00 = $60.00 to the creator. Type your own figures to see the split change.
Payouts run through the parent
Minors can't hold ad-network accounts directly, so a teen's earnings are paid into the parent-managed account linked to their EyePost account. The parent releases the money to the teen; the teen sees the full ledger — ad revenue, running costs, markup, earning — in their own Videos tab, so the split is never a black box to either of them. Ads themselves are contextual (matched to the video, not the teen): no behavioural profiling of minors, which several countries in the compliance plan ban outright.
What this prototype shows — and what it doesn't
- Shown: the Videos area with fictional teen creators and illustrated thumbnails, the upload + simulated pre-post scan with a working state picker, the three outcomes, the parent review queue, and the earnings calculator.
- Not built: real video upload, storage, playback, or a real video scanner — those are production work on the managed stack described in How it scales. The state rules listed are plain-language design examples; the production rulebook is written per state from the law catalog, under the same place-by-place gate as the rollout.
Learning — tutor, lessons, whiteboard, and writing help
Five learning tools that live inside the teen's account. Everything on this page works in the prototype: the tutor steps through real problems, the whiteboard really draws, and the writing scanner really checks your text. These tools help a teen with their own work — the writing scanner here is a learning helper, completely separate from the safety scanner that watches conversations for grooming patterns.
1 · Math tutor AI — step by step, never just the answer
Pick a problem — or type your own — and the tutor works it with you: one step at a time, a hint first, you answer, it checks and explains, then the next step. The final answer only appears when you've walked the steps to it.
Prototype note: this demo tutor runs on built-in step logic for the problem types below. The production version uses a full AI model — the best-fit ("optimal") LLM for tutoring — so it can handle any problem a teen brings, in this same hint-first style.
TutorHi Maya! Pick a problem above — or type your own — and we'll take it one step at a time. I won't just hand you the answer; you'll get there yourself, faster than you think.
No problem started yet. Answers can be numbers like 8 or 0.15.
2 · Live lesson — one account teaching a linked child account
A lesson runs from a parent/tutor account directly into the linked teen's account. Rule: lessons only exist between accounts that are already linked under the parent-verification system — a lesson can never be started with, or joined by, anyone outside that link. Inside the session, both sides share the whiteboard below and the math tutor above.
No lesson is running right now.
Start one here to demo it — this button plays the parent's side (Dana). In the real flow, the parent starts it from the Parent dashboard and it appears in Maya's app under Learn.
Start one here to demo it — this button plays the parent's side (Dana). In the real flow, the parent starts it from the Parent dashboard and it appears in Maya's app under Learn.
Lesson live
Math — solving linear equations
Dana Whitfield (parent / tutor account) → Maya Whitfield (linked teen account). Maya's app shows this lesson under Learn, with a Join button — she has joined this demo session.
Shared only inside this session, between these two linked accounts. Nothing on the board is posted, public, or visible to anyone else.
3 · Virtual whiteboard — both sides can draw
In a live lesson, the tutor and the teen can both draw and write on this board at the same time — working a problem out together. It really draws: use your mouse or finger. (In this prototype both sides share the one board on this page; the production version syncs the same board live between the two accounts.)
Ending the lesson closes the session for both accounts; the board is wiped.
4 · English grammar cheat sheet — the quick-glance rules
The mix-ups that catch everyone, each on a card: the rule in one line, a right example, and the wrong version crossed out. Teens can open this while writing a post, a message, or homework.
their · there · they're
✓ Their dog is loud. / Over there! / They're here.
✗ Their is a dog over they're.
their = it belongs to them. there = a place (or "there is"). they're = they are — if "they are" fits, use they're.
your · you're
✓ Your coat. / You're late.
✗ Your going to love you're gift.
your = belongs to you. you're = you are. "You're welcome" — never "your welcome."
its · it's
✓ The cat licked its paw. / It's raining.
✗ The dog wagged it's tail.
its = belongs to it (no apostrophe — like his/hers). it's = it is, every time.
to · too · two
✓ Go to school. / Me too! / Two dogs.
✗ I want to go too the store, to.
to = towards, or part of a verb (to run). too = also, or more than enough (too much). two = the number 2.
then · than
✓ First lunch, then class. / Taller than me.
✗ Better then yesterday.
then = time — what happened next. than = comparing — bigger than, more than, rather than.
affect · effect
✓ Rain will affect the game. / The effect was huge.
✗ The affect of the rain was big.
Memory trick — RAVEN: Affect is a Verb, Effect is a Noun (most of the time).
loose · lose
✓ These jeans are loose. / Don't lose your keys.
✗ We're going to loose the game.
loose = not tight (it rhymes with goose). lose = the opposite of win, or to misplace something.
Subject–verb agreement
✓ She runs. / They run. / The team is ready.
✗ She run fast. / The team are ready.
One person/thing → verb usually takes an -s (she runs, it is). More than one → no -s (they run, we are). A group acting as one unit (the team) counts as one.
Apostrophes
✓ Maya's bag. / The dogs' bowls. / Don't, can't.
✗ The dog's are barking. / Its mine.
Apostrophes show possession (Maya's) or missing letters in contractions (don't = do not). Never use one just to make a plural: "two dogs," not "two dog's."
Punctuation basics
✓ "We won," she said. / It's over — finally!
✗ We won she said the game was over
End every sentence (. ! ?). Commas separate lists and pause clauses: "After the game, we ate." Capitalise names and the start of sentences — and the word I, always.
5 · Writing scanner — your words, checked and explained
Paste or write anything that's yours — homework, a post draft, an essay paragraph. The scanner catches commonly misspelled words and right-spelling-wrong-word mistakes (then/than, your/you're…), suggests the fix, and explains why it's correct — teaching, not just autocorrect. Click a suggestion to apply it in your text.
Separate from the safety scanner: this tool reads text you choose to paste here, to help you write. It has nothing to do with the AI safety scanning of conversations described in the Terms of Use.
Prototype dictionary: a built-in list of common misspellings plus context rules for the classic wrong-word pairs. The production version pairs this with the full AI model for trickier cases — and still explains every fix.
Terms of Use — the safety rules, in plain language
Every parent accepts these terms at sign-up, and the message-scanning policy below is stated the same way in three places: here, during parental sign-up, and in the parent dashboard. Nothing about scanning is hidden or buried.
1. Who can use EyePost
EyePost is for teens aged 13–17 only. There is no teen-only sign-up: a parent or legal guardian must first create an account, pass an identity check (government ID plus a live selfie that matches it), and then create and approve the teen's account. A teen account stays inactive until its linked parent approves it.
2. Message scanning — AI only, no human eyes
By using EyePost, parents and teens agree to the following, which is a core condition of the service:
- All conversations are scanned by AI by default. The scanner looks for patterns linked to grooming and exploitation — requests for secrecy, gifts or money offers, asking for photos, pressure to move the conversation to another app, fishing for personal details, and plans to meet or travel.
- No human reads your teen's messages. Scanning is performed by software only. No employee, moderator, or contractor of the app reads teen conversations — flagged or otherwise.
- If a threat is detected, the parent is alerted. A high-risk conversation is paused automatically and the linked parent is notified through the parent dashboard.
3. What parents can see — flagged messages only
Parents monitor their own child through the parent dashboard. When a conversation is flagged, the parent can review the flagged messages there and decide whether the contents are appropriate for their child.
Parents can only access messages that get flagged. Unflagged conversations are never shown to parents, and the app itself never displays a teen's routine messages to anyone. There is no way in the parent dashboard to browse a teen's chats, posts, or unflagged activity.
4. If something is wrong — police and the tip line
Reviewing flagged content is the parent's decision to make. If a parent decides the content is not appropriate — or believes a crime may be involved — they can call the police.
The app provides the parent with a direct link to the national tip line with every high-risk alert: the NCMEC CyberTipline (report.cybertip.org), where suspected child exploitation can be reported. Flagged messages are preserved in the dashboard so a parent can share them with police or the CyberTipline.
5. Identity verification and face data
Parent identity checks are performed by a specialist verification provider. The app stores only a "parent verified" result and the provider's check reference — it never stores ID images or face data in its own database.
6. The encryption trade-off, stated plainly
Because messages are AI-scanned for safety, they are not end-to-end encrypted. This is disclosed here, at parental sign-up, and in the parent dashboard — not hidden in fine print. Families who do not accept AI scanning cannot use the messaging part of EyePost.
7. Where EyePost is available — country by country, state by state
EyePost opens place by place: US states one at a time first, then countries one at a time. A place becomes available only after its laws have been cataloged for review — age floor, parental-consent rules, ID and biometric rules, child-safety duties, and the local reporting path, each with its source law named — and the app has been configured to match that catalog. No lawyer sign-off is part of the gate; the completed, reviewed catalog is the gate. Age floors follow the state or country the account is registered in, not one global rule: for example, Australia is 16–17 only, because its under-16 social-media ban cannot be overridden by parental consent.
North Korea, China, and Russia are not served. Access from those countries is geo-blocked: sanctions (North Korea), and data-localization and state-access laws (China, Russia) that conflict with this service's model — AI-only scanning, and parents able to see flagged messages only.
Every parent alert includes the local tip line or reporting route for the account's country, not only the US CyberTipline. One legal duty applies in the US: if the AI flags apparent child sexual abuse material or enticement, federal law (18 U.S.C. § 2258A and the REPORT Act) requires the service to report it to the CyberTipline and preserve the flagged content for one year — a narrow, logged escalation path that exists only for that duty.
8. Videos — pre-post scanning and creator earnings
Every video is scanned before it can post. Nothing is published first and checked later. The scan reads the video's frames, audio, and on-screen text and checks them against (a) a federal baseline that applies in every state — including sexual content involving a minor, which is always blocked and, where federal law requires, reported to the NCMEC CyberTipline with the content preserved — and (b) the rules of the state the poster's account is registered in.
- Approved videos post normally.
- Held videos do not post. They go to the linked parent's dashboard, and only the parent can approve or keep them blocked — the same flagged-only rule as messages: parents never see a teen's approved videos in the dashboard.
- Blocked videos never post and cannot be approved. The parent is notified.
Creator earnings follow John's formula: creator earning = ad revenue − cost to run − 10% markup. From the ad revenue a teen's videos bring in, the service subtracts what those videos cost to run (hosting, streaming, and scanning), then subtracts a markup of 10% of the ad revenue; the remainder is the teen's earning. Because minors cannot hold ad accounts directly, earnings are paid to the parent-managed account linked to the teen's account, and the full split is visible to both the teen and the parent. Ads are matched to the video's content, never profiled to the teen.
9. Learning tools — tutor, lessons, and the writing scanner
EyePost includes learning tools inside the teen's account: an AI math tutor that works problems step by step, live lessons, a shared whiteboard, a grammar cheat sheet, and a writing scanner.
- Lessons run only between linked accounts. A lesson can be started only by the verified parent/tutor account linked to the teen, and only that teen can join it. The whiteboard is shared only inside the lesson session.
- The writing scanner is a learning helper for the teen's own text — it checks writing the teen chooses to paste in, flags common misspellings and wrong-word use, and explains each fix. It is a separate tool from the safety scanning in section 2, which looks for grooming patterns in conversations.
- The math tutor teaches; it isn't an answer machine. It gives hints and steps first and checks the teen's own answers along the way.
Prototype note: this is a plain-language design draft of the terms for the prototype, a design draft — each place's live terms come from its reviewed law-catalog entry.