KinBot is KinShield's free scam checker. Paste a text or an email, drop in a screenshot, or share a voicemail. It points to the exact words that look like a scam, checks any links and phone numbers, then tells you what to do next.
Live on Amazon Bedrock · gpt-oss-20b, Qwen3-VL, Mistral Voxtral · no sign-up needed · your message is never stored
A real result from the live app. Run this example yourself; it takes a few seconds.
Paste a text or email (up to 2,000 characters), add a screenshot, or share a voicemail. For screenshots, a vision model on Amazon Bedrock (Qwen3-VL) reads the words and looks at logos, links and QR codes. For voicemails, a speech model on Bedrock (Mistral Voxtral) writes down what was said; the audio isn't stored.
KinShield's evidence detector (gpt-oss-20b on Amazon Bedrock) looks for seven fixed scam signs. It must quote the exact words, and KinBot drops any quote that isn't really in your message.
After every verdict, an agent (gpt-oss-20b with tools) checks how old a website is and whether it imitates a brand, looks it up in phishing feeds, opens the link safely as a plain static fetch with no browser, checks phone numbers, and pulls FTC-based advice.
Plain steps, like calling back on a number you already have. Then ask follow-ups, like "but it came from his real number?"
The same fixed list KinShield uses on phone calls. Kip can't invent new ones, and every sign it reports comes with the words that raised it.
Grandma, it's me.
I'm at the police station.
Please don't tell Mom.
I need $2,000 today.
Pay with gift cards or Bitcoin.
This is Officer Daniels with Social Security.
Your account locks in 30 minutes.
Your message, screenshot or voicemail is checked and forgotten, and the conversation lives only in your browser tab. Signing in is optional. If you do, Kip saves the verdict and headline of each check (never the message itself) so you can see your history.
KinBot is an AI demo. A LOW result doesn't prove a message is safe. If money is involved, call back on a number you already trust.
KinModel, our tiny keyword model, adds a second opinion. It's clearly labelled, and Kip's cited answer always comes first.
Kip only reads what you share with it. It never sees your calls, contacts or accounts, and it will never ask for a password or code.
Honest answers about what KinBot does, and what it doesn't.
Yes. No sign-up is needed. Signing in is optional. If you do, Kip saves the verdict and headline of each check (never the message itself) so you can see your history. KinShield is pre-launch, with no paying customers.
Text up to 2,000 characters (a text message, an email, or what a caller said), a screenshot, or a voicemail recording. Remove names and account details first.
All on Amazon Bedrock. The evidence detector is the same one as the KinVoice call demo: OpenAI's open-weight gpt-oss-20b with a structured prompt and server-side checks. Screenshots are read by Qwen3-VL, voicemails are written down by Mistral Voxtral, and the link investigator is gpt-oss-20b with tools. The KinModel second opinion is a small TF-IDF + logistic-regression classifier that runs inside the same AWS Lambda.
Only safely. The investigator fetches the page from AWS as plain static text (no browser, no scripts, no cookies, no forms), strips tracking codes from the link, and only keeps features like the page title and forms, never the page itself. It also checks the domain's age, brand look-alikes and public phishing feeds.
Call your bank or payment provider now, using its official app or the number on your card, and report it at reportfraud.ftc.gov or ic3.gov. Tell KinBot "I already sent money" for the full checklist.
Open KinBot, paste it or add a screenshot, and Kip will show you the exact words that worry it. No sign-up needed.
