Product Predict
v0.5 · local CLI · public demo
User Experience Simulation · not a test runner
Find out how strangers feel
about your product before they meet it.
Product Predict generates a population of synthetic users tailored to your product — each with their own preferences, prior tools, and competitor experience — and lets them use it in a real browser. They quit when a real user would quit. You get back not a bug list, but how that population felt about your product: the design issues, the misalignments, the it-doesn't-fit-this-kind-of-person you couldn't see yourself.
How it works
01
AI generates the population
Given a one-liner about your product, pp produces 5-10 personas with preferences, prior tools (incl. competitors), and weights summing to a realistic distribution. Edit them as JSON. Or derive them from real beta data — docs + audio.
02
Each one uses it in a real browser
Headless Chromium, your target URL. Every persona has their own context, their own POV, their own patience. They explore what they care about. They skip what bores them. They quit when satisfied — or frustrated, or just curious-out.
03
You get a feelings report
Not a bug list — observations like ‘the navigation rhythm is off for someone coming from Asana’ or ‘the empty state doesn't sell the product to a non-technical user.’ With evidence: real screenshots, the agent's own words, where they hesitated.
Install · no signup, no API key
One command. Zero config.
pp is a local CLI. Reports stay on your laptop. The simulation pool runs on our side — you get a free quota by virtue of having the binary installed. No accounts, no tokens.
install — one line
$ curl -fsSL https://product-predict.vercel.app/install | sh· Clones to ~/.pp/src, symlinks pp into ~/.local/bin
· Playwright + Chromium auto-downloaded on first install (~150 MB, one-time)
· Re-run anytime to upgrade — it git-pulls the latest
· Requires git, node ≥ 20, npm
· Playwright + Chromium auto-downloaded on first install (~150 MB, one-time)
· Re-run anytime to upgrade — it git-pulls the latest
· Requires git, node ≥ 20, npm
first run
$ pp run https://your-app.com --hint "team todo app"
AI generates ~6 personas tailored to that description, drives them through your URL in real Chromium, writes ./runs/run-001/report.html. ~3 min.
See a real report
run-006 · target: localhost:8908 (a 4-tab todo demo)
7 agents from 5 personas. 5 quit frustrated · 2 quit explored. 14 experience observations across 10 categories — most were not bugs but design/fit/competitor mismatches. Feature usage chart shows where users actually spent attention.
王磊 (Asana 3-year user): "Trello at least shows me cards. This thing just makes my task vanish."