Anyone can flood an endpoint with identical requests. Production traffic doesn't look like that. AgentAPI gives you a fleet of independent, scripted agents that drift, spike, misbehave and retry - like your real clients do. Then it measures everything.
Generating load is the easy half. AgentAPI's job is the other half: telling you precisely what broke, when, for whom - while the run is still going.
true percentiles per endpoint, not averages that hide the pain
refresh on every chart - watch failures happen, not read about them after
drill from the fleet down to one client's single failing call
the whole run, raw, one click - for your own graphs and the audit trail
Every agent runs its own little program: call, wait, loop, chain values from one response into the next request. If you can write a shopping list, you can script an agent.
Gaussian noise, slow drift, hard clamps, the occasional 2% spike. Flat synthetic values never broke anything - realistic ones do.
OpenID Connect built in: one service account for the fleet, or a personal login per agent. Tokens refresh themselves mid-run so 4 a.m. soak tests don't die at 4:05.
Your own tenant: users, agents, metrics and config fully separated. SSO login, role-based access, nothing shared with anyone.
This is a battery that drifts, jitters and occasionally spikes - reporting every second. Six lines, written in the browser, validated as you type. No SDK, no build step, no YAML archaeology.
Point AgentAPI at your host, describe an agent - endpoints, payloads, behaviour - then clone it into a fleet of 350 with one click.
Launch one agent or all of them. Pause, resume, stop. Each agent is independent, with its own live panel, program trace and countdown.
Watch p50/p95/p99 and error rates move in real time, drill into any agent, any endpoint - then take the CSV home for the post-mortem.
"We built AgentAPI because our own load tests kept passing while production kept failing. The tools hammered one endpoint with perfect, identical requests - our real devices did no such thing. So we wrote a simulator where every client is a little program with its own noisy life. Our tests started failing in the right places. Then we made it a product."
Hundreds of sensors reporting on their own clocks, with their own quirks.
Will the pipeline survive a week of realistic noise, drift and spikes?
Overnight runs with auto-refreshing auth and a CSV waiting in the morning.
Tell us where to reach you and we'll set up your private workspace - your tenant, your agents, your dashboards. A real person replies, usually within a day.