Real reportOnline shop with a small catalog
Small catalogHolds17 req/s
Holds ~17 req/s on short steps (2.7 min), fails at ~29 req/s
- Breaks by
- 28.5 req/s
- Slowed first
- Add to cart
- Latency bound
- p95 ≤ 1.5 s
Load testing with numbers you can check
We run the agreed test and return the measurements, the result of your chosen rule and the limits of what was observed. You and your AI read them with your knowledge of the application.
Measure your targetA target is the address that may receive load.No subscription · the Run's price is visible before you confirm
Let your favorite AI analyze whether CapacityLab may be a good fit for your team
Why these numbers hold up
A second, independent measurement point next to your own telemetry.
Your logs stay useful. A Run adds what the client side saw, under a load you agreed in advance.
- Numbers measured at loadp95 and the error share are computed from the responses our generator received: the client-side view of the whole request path.
- You choose the targetLoad goes only to the endpoint you registered: staging by default, production by your decision.
- The price is locked before trafficYou accept the target, the workload and the exact price of the run before the first request.
Pricing
Credits in dollars, no subscription.
Every Run shows its exact dollar price before you confirm it and is paid from your Account's balance. Credits never expire.
Free
A new Account starts here with $5 of welcome credit.
- Each Run: up to 20 RPS, 50 concurrent requests and 5 minutes of traffic
Paid
After a credit pack is bought or CapacityLab grants credit.
- Each Run: up to 1,000 RPS, 1,000 concurrent requests and 720 minutes of traffic
Credit pack
$19 of Run credit for $19
Sales are closed during the beta, so there is nothing to buy yet. Pricing details.
Talk through your workload
Book a short capacity review.
Bring a target endpoint, a traffic question, or a Railway bill you do not trust yet.
One controlled loop
Load profile in. Capacity decision out.
The application stays the target. Test infrastructure stays in support.
- 01Authorize
You attach CapacityLab as a tool in your own AI agent and grant only the scope it needs. The connection runs over MCP (Model Context Protocol), an open protocol for exactly this.
- 02Profile
Your agent turns the question into a workload profile: target, endpoints, traffic shape and rate.
- 03Register
Register the exact endpoint that may receive load — staging by default, production only if you decide so.
- 04Run
We send the bounded workload from our generator; no request leaves it without a registered target.
- 05Watch
You follow your own infrastructure while the load runs, on your usual dashboards, and can stop the Run.
- 06Report
You get the latency measured at the load generator, the result of your chosen rule and its observation limits. Why the application behaved so is for you and your agent to judge; HTTP measurements alone do not establish an internal root cause.
Traceable by design
Evidence tied to the test you ran.
- Registered targetOur generator sends requests only to the endpoint you registered and authorized.
- Confirmed before trafficYou accept the target, the workload and the exact price of the run before any request is sent.
- Linked evidenceEvery report keeps the test parameters and measured evidence behind its result.
Registered targets only
Ready to measure your own target?
Register the endpoint that may receive load and get the capacity decision together with its evidence.