Library / SDK
Guardrails & Structure
Guidance
1Constrain 2and 3structure 4LLM 5generation 6programmatically
the six Ws · specification
W1
Who
Maintained by the guidance-ai project (roots at Microsoft).
W2
What
A paradigm for constraining and interleaving LLM generation with control flow, regex, and grammars.
W3
Where
A Python library driving local or API-based models.
W4
When
Reach for it when you need reliable structured output or token-level control.
W5
Why
Enforces output structure and reduces wasted tokens by constraining generation.
W6
With
Works with Transformers, llama.cpp, and OpenAI-compatible APIs.
W7
Watch
Maintained by the guidance-ai community; open-source (MIT/Apache-2.0), self-hostable, no hosted service. With local models it runs entirely on your machine.