TL;DR — Guidance is a control language for constraining LLM generation with templates, regex, and token-level steering.
What it is
It lets you mix prompt text with program logic to enforce output structure.
Why it exists
When strict output shape matters, Guidance reduces parsing failures and post-processing hacks.
Install
pip install guidance
Basic usage
import guidance
# define structured generation template
# run and parse constrained output
When to use, when to skip
Use it when this category is a bottleneck in your agent stack and you want faster delivery with fewer custom components.
Skip it when your workload is tiny, requirements are fixed, or a plain provider SDK plus a few local functions is enough.
Alternatives
Compare with adjacent tools in the same AI Native category and choose based on interface style, deployment model (hosted vs self-hosted), and team familiarity.
Verified against project documentation, June 2026.