pydantic-ai-testing

Use TestModel to run deterministic tests of PydanticAI agents without calling external APIs. TestModel replaces live LLM responses with controlled outputs, enabling reproducible unit tests and eliminating API dependencies during development.

Yes. FunctionModel and TestModel allow you to mock LLM responses and force specific tool calls, giving you complete control over agent behavior during testing. This lets you validate tool invocation logic independently of LLM output.

Use VCR cassettes to record API interactions during test runs and replay them on subsequent runs. This approach captures real responses once, then replays them deterministically without repeated API calls, integrating with pytest test infrastructure.

Combine TestModel with inline snapshots and seed control to generate and assert deterministic agent outputs. Inline snapshots capture expected outputs inline in test code, and controlled seeds ensure reproducible results across test runs.

Yes. PydanticAI testing patterns support dependency overrides, letting you inject mock implementations of external services and databases during tests. This isolates agent logic from infrastructure and ensures tests remain fast and deterministic.