AgentrekAgent simulationJourney replay

Prove agents can finish the journey, not just find you.

Discovery is table stakes. Agentrek runs real agents and agent swarms against your infrastructure to show which journeys complete, which break, and the evidence for both.

Simulate. Observe. Measure. Prove. Four verbs, in that order, and nothing asserted without the fourth.

Run an Agentrek See how it works

01 / The gap

Your agents are only as capable as the infrastructure they touch.

Websites were designed around a human reading a screen. An agent has to discover, understand, navigate, authenticate, act, recover from its own errors and finish the task. Smart agents, unprepared infrastructure. Agentrek closes the gap by putting agents on it.

Traditional testing asks

Does the website work for humans?

Agentrek asks

Can an agent actually use it?

02 / What it is

An agent simulation and testing environment for the web.

Agentrek deploys actual agents, alone and in swarms, against your website and digital infrastructure, then records what they did rather than what they were supposed to do. Every run is recorded end to end: the action trace, the agent's own reasoning at each step, and the screen it was looking at when it decided.

Four agents descending the same route in single file — One route, many agents. A trek is the same journey run again and again, by agents that do not all behave the same way.

Define

The journeys that carry value, and what finishing one actually means.

Simulate

Real agents and swarms, with different goals and strategies.

Observe

Actions, reasoning, screens and recovery, recorded as they happen.

Analyse

Where the journey held, where it broke, and what caused it.

Evidence

The trace attached to every finding, reproducible by your team.

Captured on every run

ten records per step, kept for replay

✓01

Agent actions

✓02

Agent reasoning

✓03

Screen state

✓04

Agent responses

✓05

Errors and recovery

✓06

Task completion

✓07

Path taken

✓08

Time and performance

✓09

Token cost

✓10

Final outcome

03 / Evidence

Don't trust a score. See the evidence.

Most assessments tell you what they think is wrong. Agentrek shows you what happened. Every finding traces back to an observed agent behaviour, and nothing reaches a report without the trace attached.

Agent attempts the journey

Page state captured

Action, reasoning and response recorded

Failure identified

Evidence attached to the finding

04 / What it tests

Test your infrastructure the way agents experience it.

Simulation

Agent simulation

Deploy realistic agents against your website and digital infrastructure.

Scale

Agent swarms

Many agents with different behaviours, goals and strategies, to surface edge cases.

Journeys

Real digital journeys

Whole workflows end to end, not a check that pages load.

Interaction

Browser interaction

How agents discover, navigate, click, input and read your interface.

Resilience

Error recovery

What happens when things go wrong, and whether the agent can recover.

Capture

Evidence capture

The actions, states, responses and outcomes behind every result.

Scoring

Agent evaluation

Whether the task completed correctly, efficiently and repeatably.

Replay

Agent journey replay

We record the full journey, so you can replay the exact sequence and read the reasoning behind the step that failed.

05 / How it fits

Know what your infrastructure exposes, then prove agents can use it.

Agentrek is not another website score. The deterministic readiness layer measures what your property exposes. Agentrek validates behaviour against it. Evidence is what makes either one defensible.

Measure

Webzero readiness

The deterministic signals: what an agent is permitted to see, understand and call. Reported as the w0 score.

Validate

Agentrek

Real agents and swarms run the journey, so behaviour is observed instead of inferred from the signals.

Prove

Evidence

Actions, reasoning, screens and outcomes, attached to each finding and replayable by your engineers.

06 / Example run

Watch an agent test your infrastructure.

Every run produces a complete evidence trail, from the first action to the final outcome.

AGENTREK / RUN VIEWtrk-4471 · agent A-03 · property redacted41 actions6m 12sstopped

Task

Open a fixed deposit for 18 months, funded from the linked savings account, and retrieve the confirmation document.

Agent response

Reached the deposit form and set the tenure, then could not complete funding. The account selector exposed no accessible value and the confirmation step required an OTP with no programmatic path. Retried three times, then stopped.

Session recording

video capture on

00:12 / 06:124 markers

Action trace

4 of 41 actions shown

01NAVIGATEfound the fixed deposit page✓

02INPUTset the tenure to 18 months✓

03READfunding account list, no readable value!

04STOPotp step, no path the agent can complete✕

Recorded reasoning, step 04

I cannot read a value from the funding account list, so I cannot confirm which account to draw from. The OTP step has no path I can complete. Three retries gave the same result, so I am stopping rather than submitting an unverified deposit.

Evaluation

✓Found the journey on its own

!Tried three times to recover

✕Did not reach the outcome it was sent for

Metrics

actions

retries

6m 12s

time on task

$0.38

token cost

Evidence attached

session videojourney replayreasoning tracescreen state × 12response log

Journey✓Find the product✓Sign in!Fund the deposit✕Confirm

Illustrative run. Client identity, domain and trajectory data are never published, including on this site.

07 / Why Agentrek

Stop testing whether your website works. Test whether agents can use it.

Traditional testing

Agentrek

Tests predefined functionality

Tests how agents actually behave

Human-centric

Agent-centric

Mostly pass or fail

Behaviour plus evidence

Limited edge cases

Agents and agent swarms

Reports the result

Shows why

Static tests

Real interaction

Assumes the journey

Observes the journey