Sniff — The AI Mystery Shopper for Your Website

The Problem: Every founder eventually asks the same question: "Why aren't people signing up?" The honest answer is usually buried in analytics tools that show what happened (a 40% drop-off on step 3) but never why. Watching real user session recordings doesn't scale, and traditional QA testing only checks whether a flow technically works — not whether a confused, impatient, or first-time visitor can actually get through it. Landing page audit services exist (we studied existing solutions), but they're expensive per-report products that paywall most of their own findings. The Solution: Sniff is an autonomous AI agent that behaves like a real visitor — not a scripted test runner. Point it at a signup flow or a landing page, and it drives a real browser (Playwright) step by step, deciding what to click based on what it actually sees on screen, the same way a person would. It doesn't follow a hardcoded path; a vision-capable model (Gemini) looks at each screenshot and decides the next action, while a separate fast-reasoning tier (Jev) sanity-checks those decisions and catches ambiguous "did we actually succeed?" moments a keyword search would miss. Two products in one agent: 1. Signup/onboarding testing — assign the agent a goal ("sign up for an account") and a persona (confused first-timer, impatient power user, careful reader), and it attempts the flow, then classifies exactly why it failed if it did: root cause, severity (P0–P3), and likely owner (Backend/UX/Performance/Integration), escalated via Slack. 2. Landing-page conversion audit — point it at any URL and it produces a full product-manager-style teardown: an overall score, a five-dimension breakdown (Message & Clarity, Audience Fit, Action Path, Trust & Credibility, Content Depth), an annotated screenshot showing exactly what it identified as the CTA and clicked, a step-by-step narrative of what it saw and concluded, dominant colors/fonts actually sampled from the page, broken-link checks, copy rewrites, and a prioritized fix list. Every field a paid competitor report locks behind a paywall, Sniff shows in full. Who it's for: founders and PMs who want an honest, evidence-backed answer to "what's wrong with my funnel" without hiring a researcher or paying per report; and engineering teams who want signup-flow regressions caught automatically, with a diagnosis attached instead of just a red X. How they interact with it: a web dashboard — paste a URL, pick a goal or audit mode, and watch the agent work in near-real-time, then read the report with the same screenshots and reasoning trail the agent used. What makes it creative and unique: most "AI testing" tools still follow scripted selectors. Sniff's agent reasons from pixels and text like a human tester, deployed with a genuinely three-tier architecture (vision LLM, fast text LLM, ultra-fast decision model) that keeps it both smart and cheap to run at scale — and it's the only tool in this space that treats conversion auditing and functional QA as the same underlying capability: an agent that understands what it's looking at.