GU

Guidance

Constrained generation language for controlling LLMs

Self-Hosted AI ★ 21.7k stars Medium setup MIT

Guidance is a programming paradigm from Microsoft for steering language model output with interleaved generation, control flow, and constraints. It can run against locally hosted models to produce reliable structured results.

Key features

  • Interleaved generation control
  • Token healing
  • Regex and grammar constraints
  • Works with local models

Strengths

  • Released under the MIT license
  • Mature project with 21.7k GitHub stars
  • Written in Jupyter Notebook

Guidance replaces

Compare Guidance

30 head-to-head comparisons.

Similar self-hosted ai apps