APT Performance Assistant

Agent Performance Testing (APT) reports on a whole test run through APT - Analysis. The Performance Assistant works at a smaller scale. It answers one question about a result, or makes one change to a test, in the AI Chat panel.

There's nothing to open or activate. Enter your request in AI Chat. APT then builds an assistant for that request alone, with only the skills it needs. A narrow assistant reads your results more accurately than one that carries every capability at once.

Enter a request

Name the result, test, transaction, or time range you want APT to work on. APT handles the following kinds of request:

Request

Examples

Investigate a result

"Why is the checkout transaction slow?"

"What happened around 00:30?"

"Is a 2.1s login actually a problem?"

Compare two runs

"Compare the payment transaction between today's run and last week's."

Run and configure tests

"Run the Black Friday scenario in the EU zone."

"Stop the test that's running now."

Check trends and set objectives

"How is this test trending across runs?"

"Track login p95 and fail it over 800ms."

Check and change zones

"Which zones can I use for this test?"

"Move the checkout test to a cloud zone."

Build dashboard tiles

"Build a tile for p95 of my top five transactions."

Start a deep dive from a result row

Element names in a result run long, and they often differ by a few characters between user paths. Typing one into AI Chat costs you time, and a single wrong character sends APT to the wrong element. Start from the row instead, and APT works on the element you selected.

To start a deep dive, follow these steps:

  1. Open a test result, then go to the Values tab.

  2. Right-click an element row, such as a transaction, page, or request, and select Deep dive with APT. Alternatively, select the row and open the More options menu.

  3. Check the request that APT composes, then select Send.

APT opens AI Chat and answers with a deep dive into that element.

AI Chat offers the same deep dive as a starting point under Test result overview. Select Deep dive into an element to compose the request from the panel instead of from a row.

The element stays in scope for the rest of the conversation, so your follow-up questions reach it without naming it again. Start a deep dive on another row to move the scope to that element, or clear the scope to ask about the whole result again. The screen context that APT picks up on its own stays in place either way.

Approve a change before it runs

Requests that only read your data run straight away. APT asks first before it runs or stops a test, changes the load, or reassigns zones. It shows a plain-language summary of the change: which test it affects, what changes, and what it changes to. Select Approve to continue, or Deny to cancel.

APT doesn't ask when it only gathers information. Before a zone change, it reads your zone list to check that the zone exists. It then asks you to approve the change itself.

Check the evidence behind an answer

APT reports the p95 and p99 percentiles next to the average, so the slowest requests stay visible. It judges a metric against your baseline or Service Level Agreement, and tells you when you have neither.

Every cause APT names comes with the metric, timestamp, or error pattern behind it. When your request leaves out a detail, such as which test to run, APT asks you for it.

APT reads your setup the same way. When your test points at a zone with no controller or load generator, APT tells you the test can't run. It then lists the zones you can use instead.

Know the limits

A report on a whole test run comes from an APT - Analysis, not from the Performance Assistant. APT reaches only the workspaces, tests, and results that your own permissions allow. Each request uses AI credits in proportion to the amount of information APT processes.

Deep dives have the following limits:

  • The message box doesn't accept @ references. Attach elements from the row menu or from the Deep dive into an element starting point.

  • A deep dive reaches only the elements on your current screen. To analyze an element in another result, open that result first.

What's next

Open AI Chat to enter your first request. Check workspaces and usage to see who has access and how much quota you've used. Our AI data usage policy covers which data APT processes and where it goes.