Nelli

Nelli (Nahuatl: "truth") is a fraud-verification agent for people in Mexico and Latin America who receive suspicious messages, posts or reels. It was designed for the hardest case, a 60-70-year-old who fears breaking something, so anyone can use it alone. You paste a text, upload a screenshot, send a voice note or share a link, and Nelli returns a plain-Spanish card: a four-level risk label, the evidence in 5 points, one single step to take, the verified official source, and Report / Block / "Ask someone you trust" / "I already fell for it" / "Shield my account in 7 minutes". Rules decide; the AI explains. Deterministic rules checked against a hand-verified Mexican package (official domains, what each institution publicly says it will never ask, known scam patterns) set the risk level. Muse Spark 1.3 (Meta Model API, Standard mode) only reads, transcribes and explains: it can raise an alert, never lower it, and "Confirmed fraud" only comes from hard evidence that does not depend on the model. Nelli never opens links, stores nothing and never says "it's safe". Built with the Meta Model API (Muse Spark 1.3, Muse Voice Transcribe, Muse Image) and open-weight Prompt Guard 2. Unaffiliated prototype, not a Meta product.