Guidance

Overview

Guidance is a framework for steering large language model outputs through a domain-specific language that constrains token generation. It lets you specify exact output formats, control branching logic, and enforce structured responses without post-processing.

Best for

Best for
Developers building production systems that need deterministic, schema-compliant LLM outputs

Use cases

• Enforce JSON or XML schema compliance in model outputs
• Build multi-turn workflows with conditional branching based on model responses
• Extract structured data from unstructured text with guaranteed format

Notes

Guidance is a framework for steering large language model outputs through a domain-specific language that constrains token generation. It lets you specify exact output formats, control branching logic, and enforce structured responses without post-processing.

21,486 stars on GitHub. Last updated 2026-05-21. Licensed MIT.

Use cases

• Enforce JSON or XML schema compliance in model outputs
• Build multi-turn workflows with conditional branching based on model responses
• Extract structured data from unstructured text with guaranteed format

Pros

• Reduces hallucination and invalid outputs by constraining generation at token level
• Eliminates need for output parsing and validation in downstream code
• Works across multiple LLM providers and local models

Cons

• Adds latency due to constraint checking on every token
• Requires learning a new DSL syntax for non-trivial use cases
• Community-maintained with no commercial support guarantee

Indexed from awesome-llm and enriched against its public facts.