AI Trust Centre

The AI Trust Centre is where you find out whether your AI agents do their job before your customers do. The Playground tells you whether one conversation works today. The Trust Centre keeps a suite of tests you can run again after every change, so you know the agent still keeps its rules, uses its tools at the right moment, and hands over when it should - and you know what to fix next when it doesn't.

In the studio it's labelled AI Trust Center, with the US spelling.

New to it? Get started covers what to test, how to read a run, and how to keep your tests worth trusting.

Nexus Tip

Ask the Nexus AI where your suite is thin. It lists your agents with few or no tests and suggests what to add for each.

Ask the AI layerWhich agents have no tests?

How you use it

1. Try a change in the Playground until the agent behaves.
2. Capture what matters as tests in the Testing Lab: the agent's main job, each of its rules, and every real conversation that went wrong.
3. Run the agent's tests after every change to its instructions, tools, model, routing or knowledge.
4. Read the results, decide whether the agent or the test is wrong, fix it, and run again.

Use the Playground to try; use the Trust Centre to check and to decide what to fix.

Where things are

In the studio	Use it when
AI Trust Center → Testing lab	You want to create tests, group them into datasets, run them, or look back at past runs.
AI Trust Center → Evaluators & rules	You want to change what counts as a good reply on every run.
AI Trust Center → User profiles	You want tests to run as a particular kind of customer - a signed-in member, a guest, an account in another country.
The Tests button on an agent's page	You're editing an agent and want to see how its tests are doing, or run them, without leaving it.

Pages in this section

Page	What it covers
Get started	Build a suite that covers your agent, read what a run tells you, and keep your tests trustworthy.
Testing Lab	Find, organise and act on your tests and datasets.
Create test cases	The ways to create a test, the preview Nexus runs before saving one, and the checks you can add.
Run tests	Start a run, test as several customers, repeat tests to spot inconsistent results, and use mocks.
Agent rules	See which of an agent's own rules it followed or broke in a run.
Evaluators & Rules	The scores every run uses, and how to tune them.
Test Case	One test in detail: its conversation, what the agent did, its checks, and its recent runs.
Reports	Read a finished run: what passed, what failed, and why.
Test signals on the agent page	What the agent's Tests button tells you while you edit.
Execution traces	The step-by-step record of what the agent did on each turn.
Debugging	Turn a failing test into a fix.

Ask the Nexus AI

The Nexus AI panel can read your tests and runs, explain failures, draft new tests, and start runs for you.

Nexus Tip

Nexus finds your most recent failed run and explains, in plain words, what went wrong in each failed test.

Ask the AI layerWhy did my last run fail?

Nexus Tip

Turn a run with many failures into a short list of causes - an instruction, a missing branch, a knowledge gap, a tool - so you fix the cause once instead of each test.

Ask the AI layerGroup failures by root cause

Nexus Tip

Have Nexus read an agent's goal and its existing tests, then propose the tests that would close the gaps. You review them before anything is created.

Ask the AI layerSuggest new test cases for an agent

Read next

• Get started - if you're building a suite for the first time.
• Create test cases - if you know what you want to test.
• Reports - if you have a run to read.
• Debugging - if a test is failing and you don't know why.