DialogAssert documentation

DialogAssert is a behavioural QA runner for AI agents. It tests whether an agent behaved correctly in realistic multi-turn conversations, especially in regulated workflows such as banking support.

Scope

DialogAssert does not certify that an agent is compliant or lawful. It produces evidence-backed findings, in the same way a linter or a security scanner does.

The product loop

1. A business user writes a customer journey and acceptance criteria.
2. Engineering points DialogAssert at the agent through an adapter config.
3. DialogAssert runs the journey against the real agent.
4. DialogAssert captures transcript, tool calls, tool results and final output.
5. DialogAssert evaluates the run against workflow expectations, acceptance criteria, optional trajectory checks and optional deterministic assertions.
6. DialogAssert writes JSON and HTML reports with evidence, findings, scores and reproduction commands.

Where to start

• New to DialogAssert? Read the overview, then run the quickstart with no API key.
• Writing a test? Start with business journeys.
• Connecting an agent? See generic HTTP, OpenAI Agents SDK or LangChain.
• Reading results? See reading a report or browse the public sample reports.