Your product's first user is an agent.
liluser runs autonomous agent personas against your web app and reports what an AI agent actually experiences: can it complete the task, where it gets stuck, and which parts of your app are agent-hostile.
How it works
- Point us at a URL. Live site, staging, or a preview deploy — truly blackbox, no hooks or instrumentation. Trigger from the API, the CLI, or the GitHub Action on every PR.
- A persona drives your app. An LLM operating a real browser signs up, confirms its email, and tries to accomplish its goal — like a new user would.
- Get an oracle-verified report. Green checks where the experience was smooth, annotated findings where it wasn't — with side effects (the confirmation email actually arriving) verified, not self-reported.
Examples
marcus × TaskPuffin (our seeded-flaw demo app)
“Email verification becomes pure theater” — caught the on-screen confirmation code, plus the generic “something went wrong” validation. Every finding carries screenshot evidence.
dana × TaskPuffin (our seeded-flaw demo app)
Verified the code was displayed on-page and that no email ever arrived — the oracle checks the persona’s real inbox, so the report says “unverified” when the app cheats.
dana × liluser (yes, we run liluser on itself)
Our own self-review: GitHub-only sign-in is a dead end for non-developers, and the sample report sells better than our copy. We're fixing both.
Quickstart
export LILUSER_API_KEY=tf_… # beta keys are minted for you
npx liluser init # wire your repo (AGENTS.md snippet)
npx liluser run --url https://yourapp.dev
npx liluser audit --url https://yourapp.dev # multi-model readiness audit
Personas: first-time-operator,
skeptical-developer, impatient-evaluator.
API: POST https://api.liluser.dev/v1/runs.
Get access
Beta is invite-only while we tune personas with design partners.