testing-assistant-conversations

To test assistant conversations in CI, use a structured testing layer like AssistantTestLayer to validate memoized LLM conversations and AI workflows deterministically. It composes an AI service, tool execution, and a memoized test DB for consistent unit and integration tests.

A memoized LLM service caches AI responses to ensure deterministic test execution. You need it when validating assistant-toolkit conversations and blueprint operations in CI to prevent flaky tests caused by variable AI outputs.

Set up integration tests for scheduled triggers by using AssistantTestLayerWithTriggers alongside seeded blueprints. This allows you to validate time-based operation handlers and blueprint registry interactions within your CI workflows.

Yes, you can use Effect and vitest to validate DXOS assistant blueprints. The Skill provides patterns to register blueprints, operation handlers, and type schemas as consistent test fixtures within the Effect/vitest testing environment.

Testing AI flows with memoized conversations is limited to the recorded responses and seeded blueprint fixtures. It validates the orchestration and tool execution logic rather than the dynamic generation quality of the underlying AI model.