guidance

You constrain LLM output by applying Pythonic syntax, regex patterns, and grammars directly to the generation process. This guarantees structured outputs like valid JSON or XML by forcing the language model to adhere to specific formatting rules.

Constrained generation supports various backends including OpenAI, Anthropic, and local Transformers models. This allows you to apply fine-grained LLM control and structured output enforcement across different model environments.

Building multi-step agentic workflows is best achieved using Pythonic control flow to guide LLM behavior. This approach provides fine-grained control over responses, facilitating complex sequences and structured generation for reliable task execution.

You ensure LLM-generated email addresses match a regex pattern by applying constrained generation rules. Defining the regex within your Pythonic syntax forces the LLM to generate values that strictly adhere to the specified format.

LLM output fails to follow structured data formats when generation lacks constraints. Enforcing grammars or regex patterns through Pythonic constraints eliminates formatting errors and guarantees valid structured outputs like JSON.