DialogAssert documentation
DialogAssert is the independent acceptance layer for AI agents. The business states what an agent must and must not do; DialogAssert runs those controls against the agent you actually deploy, verifies what happened using the evidence your environment already produces, reports what it could not verify, compares the result with the approved baseline, and produces a defensible release decision.
Run your first journey in ten minutes
Section titled “Run your first journey in ten minutes”- Install the runner.
- Connect the agent and your trace source.
- Run discovery to see what this environment can prove.
- Write a journey and its acceptance criteria.
- Run it, repeated, against the real agent.
- Read the report.
The quickstart does all of this against a local sample agent, with no API key and nothing to connect.
The core objects
Section titled “The core objects”| Object | What it is |
|---|---|
| Control set | An approved, versioned set of controls with an owner and an approval record. |
| Journey | The multi-turn conversation those controls are exercised through. |
| Evidence | What was observed, and how strongly — claimed, observed, dependency-confirmed or independently verified. |
| Finding | A place where the observed behaviour and the approved expectation disagree. |
| Inconclusive | A control that could not be evaluated here. An assurance gap, not a failure. |
| Baseline | The last approved release, and the thing a candidate is compared against. |
The architectural constraint
Section titled “The architectural constraint”The system under test may teach DialogAssert how to observe it. It may never teach DialogAssert what “correct” means.
Discovery supplies identifiers — that “block card” is currently a tool
named block_debit_card. Only an approved control set can say
that blocking requires confirmation. Generated checks that use observed values absent from the
business requirement are rejected.
Where to go next
Section titled “Where to go next”- Connecting an existing agent? Adapters and trace sources.
- Writing controls? Control sets and acceptance criteria.
- Gating a release? Baseline comparison and CI.
- Reviewing us? Security and data handling.
- Want to see output first? Browse the published reports.