AgentLens runs 100+ LLM personas through your site like real visitors — clicking, hesitating, bouncing — and tells you exactly where they got lost. No participant recruiting. No waiting for traffic.
Reproducible with a seed, streamed live as it runs, and you only pay for the steps that actually completed.
"Buy a razor under $30." "Sign up for the trial." Or no goal — let the agents wander like real visitors.
Use our diverse default matrix, or upload markdown personas of your real audience. We screen out the fake ones.
Heatmaps, drop-off paths, persona-specific failures, and an exec summary your team will actually read.
The same insights you'd get from a five-week usability study, on the same day you push the deploy.
Click, confusion, attraction, repulsion — four overlapping channels, each dot annotated with the agent's exact thought. No more guessing why users clicked.
Eight dimensions of realism — confidence, device, accessibility, intent. Mix our defaults, or bring your own audience.
Top drop-offs, persona-specific failures, prioritized fixes — written in plain English, not just charted.
Drop in a markdown file describing your target users. We screen for nonsense and prompt injections before any agent reads them.
Every agent step streams to your dashboard as it happens. No five-minute black-box waits.
Click any heatmap dot or report screenshot. Zoom in. See exactly what the agent saw — and what they thought about it.
A nurse on her second 12-hour shift browses differently from a regulatory buyer with a spec sheet. We model eight dimensions of that difference — and you can write your own.
“I need to see shipping date before I scroll. If it's not visible by the second tap, I'm gone.”
“I'll open every product link in a new tab and compare before I commit. Reviews matter more than price.”
“Unlabelled buttons make me back out. If your alt text is empty, your conversion is empty.”
“I know where everything is. Don't redesign your nav without telling me — I'll fail upward.”
“I'm not buying today. But your interstitial newsletter popup just guaranteed I never come back.”
“I need PDF spec sheets and compliance docs before I add to cart. Marketing copy is noise.”
Most heatmap tools guess where users looked. We read what they thought. Every dot is sourced from an agent's reasoning trace — hover it to see the exact line that put it there.
Run two simulations of the same site with different changes. Same persona crowd, same goal. We diff the population, surface who flipped, and write the verdict line your stakeholders need — before a single real user sees the redesign.
We pin the persona seed across both runs. The same sleep-deprived NICU nurse browses v1 and v2 — apples-to-apples, not a fresh random crowd each time.
Headline first: “v2 beats v1 by +12pp on completion rate.” Then the persona buckets that flipped, the pages that improved, and the ones that got worse.
Side-by-side heatmaps with a ⊕ diff overlay — lime where the redesign helped, rose where it didn't. Plus a line-level HTML diff for the receipts.
Every run you launch gets tagged. Drop two of them into a battle and we crunch the diff in seconds. No new setup, no new pricing.
“Current production checkout — control.”
“Replaced "Add to cart" with "Get yours today".”
Plug AgentLens into Claude Desktop, Cursor, or Claude Code with one command. Your AI assistant launches sims, watches progress stream live, and summarises the report — without you ever leaving the conversation.
$ pipx install agentlens-mcp $ agentlens-mcp configure # paste your Pro key once. that's it.
Then drop one line into your editor's MCP config and your assistant gets eight new tools — launch, wait, summarise, compare, and friends.
14/20 agents complete — through MCP's progress channel. No silent ten-minute waits.Run a 20-agent sim on staging.acme.com/checkout. Focus on mobile, and tell me what to fix before Thursday.
On it — launching the run.
14 of 20 mobile users dropped at the persona-selector modal. The three things blocking shipping Thursday:
Want me to open a Linear issue with these and the screenshots?
Pay for what you ship, not what you might use. Annual saves 20%.