AskBefore

Share a suspicious SMS, email, screenshot, link or QR code. AskBefore shows the risk, why, and what to do – and lets you ask someone you trust before you click, pay or reply.

Comment

AskBefore checks a suspicious message before you act on it. Paste an SMS or email, pick a screenshot, paste a link or scan a QR code. In one screen you get:

  • Risk level: low, uncertain or high
  • Red flags: what looks wrong, with the exact text from the message as evidence
  • What to do now: concrete next steps

A low-risk result is never shown as proof that something is legitimate.

Safe Circle. One tap sends the check to someone you trust. They see the same analysis and answer: Don't do this, Call me or Verify independently first.

Scam Card. Any result can become a shareable warning, with phone numbers, emails, links, card and account numbers removed.

Who it is for

People who get "your parcel is on hold", "your bank account is locked" or "pay this invoice now" and are not sure. And the family member they usually call for help.

Why

Scams work through urgency and isolation. AskBefore adds a pause and a second person at the moment the decision is made.

Privacy

  • Screenshots are deleted right after analysis. Only the structured result is kept.
  • Personal details are masked in previews and Scam Cards.
  • Safe Circle members only see what you choose to share.
  • History and account can be deleted in the app.

Business model (RevenueCat)

  • Free: 5 checks per month, 1 Safe Circle contact, last 10 results.
  • AskBefore+: unlimited checks, up to 5 Safe Circle contacts, full history.

RevenueCat runs the subscription: paywall built from offerings, purchase, restore, entitlement askbefore_pro, and a server webhook with signature verification and event deduplication, so the backend enforces the limits.

How we built it

  • Expo SDK 57 and React Native, TypeScript strict
  • Supabase: Postgres with row-level security, private storage, Edge Functions
  • Anthropic Claude for the analysis, with schema-validated structured output
  • RevenueCat for subscriptions

Challenges

  • The message is hostile input. A scam can contain text written to manipulate the AI. The analysed content is always treated as data, never as instructions.
  • Always a usable answer. The AI output is validated against a strict schema. Over-long fields are shortened instead of throwing away a correct analysis.

What's next

  • Push notifications for Safe Circle (integration built, not yet enabled)
  • iOS

Built With

Share this project:

Updates

Submission history