Your AI assistant does the setup
The ValidAI skill recommends task flows from your goals and research notes, tags the click targets, then pushes the prototype to ValidAI or packages it as a .validai file.
Invite-only for now · Works with Claude, Cursor, Codex
Validate AI-built prototypes with real users — before you build them for real.
You built the prototype in an afternoon. ValidAI turns it into a structured user test the same day: your AI assistant sets up the task flows, testers open a link, and you see where each one got stuck, how hard every step felt, and a replay of what they did.
Invite-only for now.
Features
Everything between "it works on my machine" and "users get it."
Built for product managers and builders who prototype with AI, not for research teams.
Remote invites or in person
Send an invite link testers open in their browser, with no account to create. Or run the session side by side on your own machine.
Steps complete themselves
A step completes when the tester clicks the tagged element or the expected screen appears. Testers rate how easy each step felt on a 1–7 scale.
Session replays
Watch exactly what each tester saw and did, for remote and in-person sessions, alongside their step results and notes.
AI insights inbox
AI findings land in an inbox you track or dismiss. Bring your own Claude, OpenAI or Gemini key; Gemini's free tier works. Rule-based signals need no key at all.
Private by design
Prototypes are private to your account. Testers can't open the files outside an open session, file links expire, and replays are yours alone.
How it works
Three steps to your first test
-
Prepare with your AI assistant
In your prototype repo, ask
Prepare this prototype for ValidAIand share your goals, notes or PRD. It proposes task flows, tags what testers should click, and sends it to ValidAI, or packages it as a file you drop in. -
Share an invite
Send testers an invite link, or run the session in person. They follow the task flows in their browser, with nothing to install.
-
Review insights
See completion and ease for every step, watch replays where people struggled, and turn AI findings into tracked insights.
See exactly where each tester got stuck
Every session review lines up the task flow step by step: what completed on its own, what didn't, and how hard each step felt.
Watch the moments that matter
Replays show what the tester saw and did, so a low ease score comes with the reason behind it.
Findings in product language
The Insights page collects signals across sessions and an inbox of AI findings, ready to track or dismiss.
- Signals on drop-off, effort and abandonment, with no AI key needed
- AI synthesis with your own Claude, OpenAI or Gemini key
FAQ
Questions, answered
Is my prototype private?
Yes. Your prototypes, sessions and replays belong to your account, and other ValidAI users can't read them. Uploaded prototype files are served from a separate origin inside a sandboxed frame, so prototype code can't reach your ValidAI sign-in. Testers can only load the files through an open test session for that exact prototype version, closing the session cuts off access, and file links use signed tokens that expire. Only you can watch replays. If you test a prototype by its public URL instead of uploading it, that URL stays as public as it already was.
Which AI tools work?
Claude Code and Claude.ai / Cowork use the ValidAI plugin. Cursor, Codex and other coding assistants use an AGENTS.md file plus the validai-share command-line helper, both downloadable from ValidAI. In Claude.ai, Claude usually hands you a .validai file to drop into ValidAI. You can also skip the assistant and upload an HTML file or a prototype folder, or paste HTML.
Do I need an AI API key?
No. Step completion, ease ratings, session replays and rule-based signals all work without one. AI synthesis is optional: add your own Anthropic (Claude), OpenAI or Google Gemini key in Settings. Google AI Studio issues Gemini keys with a free tier.
What does it cost?
ValidAI is invite-only for now, and free for invited accounts: there's no credit card and no paid plan. If you turn on AI synthesis, your AI provider bills its usage to your own key, and Gemini's free tier works.
Can testers see my source code?
Testers only receive the files of the prototype version they're testing: never your repository, your other prototypes or other versions. Those files run in the tester's browser like any web page, so someone determined could inspect them with developer tools during a session. Keep secrets and API keys out of prototypes.
Find out if it works before you build it for real.
Set up your first test in minutes with the AI assistant you already use.
Invite-only for now.