baml-integration

Type-safe conversion of LLM outputs uses BAML to define structured schemas that generate type-checked client code. Create .baml definitions in baml_src/, generate the baml_client, and use it to parse and validate LLM responses into Pydantic-compatible types, eliminating manual parsing and type errors.

Organize BAML definitions in baml_src/ with clear schema and prompt separation, then generate the baml_client into a dedicated output directory. This centralized structure keeps prompt and schema management in one place, reduces duplication, and makes regeneration straightforward as requirements change.

The factory pattern combines fast rule-based matching for simple cases with BAML for complex ones, routing dynamically at runtime. Define both rule logic and BAML schemas, then use a factory to select the appropriate handler based on input characteristics, balancing speed and accuracy.

BAML generates Pydantic-compatible type definitions natively, enabling direct integration with Pydantic validation pipelines. The generated baml_client output types work seamlessly with Pydantic schemas, supporting token-efficient prompts and robust LLM output parsing without additional conversion layers.

Regenerate the baml_client from updated baml_src/ definitions using the recommended generation tool. The new client reflects schema changes immediately, ensuring type safety across your application and preventing stale type definitions from causing runtime errors.