guidance

Constrained generation controls LLM output with regex and grammars to guarantee valid JSON, XML, or code. It enforces structured formats and builds multi-step workflows using Pythonic control flow.

The best way to ensure generated data meets strict validation rules is to apply grammar and regex constraints directly during generation, enforcing formats like valid email addresses or integers before output is finalized.

Yes, this constrained generation approach works with local Transformers and llama.cpp models, as well as various backends including OpenAI and Anthropic, providing flexible backend support for LLM execution.

You build multi-step conversational agents with predictable LLM responses by using Pythonic control flow to orchestrate complex data processing pipelines and enforce structured formats at each generation step.

Downstream parsing errors occur when generating XML from LLMs because models produce unpredictable text; applying constrained generation with custom grammars guarantees the output adheres strictly to the required XML structure.

You need the Guidance framework to enforce structured LLM output because it provides the underlying mechanism to control generation with regex and grammars, ensuring valid JSON, XML, or code across various model backends.