guidance

Control LLM output syntax with regex/grammars; guarantee valid JSON/XML/code; build structured workflows.

• Use regex or grammars to enforce valid output formats (e.g., email, date, JSON).
• Build multi-step workflows with context managers and stateful functions.
• Ensure structured, valid outputs via token healing and grammar-based generation.
• Supports multiple LLM backends (OpenAI, Anthropic, llama.cpp).

• Enforce valid syntax with regex constraints (e.g., email, phone, date).
• Generate structured outputs (JSON, XML, code) using grammars.
• Automatic token healing for natural text boundaries and spacing.
• Pythonic context managers for chat-style interactions (system/user/assistant).
• Reusable guidance functions with stateful or stateless patterns.
• Reduce latency and invalid output risk with constrained generation.

1. 1.Install guidance via pip (basic or with backend support).
2. 2.Use context managers (system/user/assistant) to structure prompts.
3. 3.Apply regex constraints to enforce valid formats (e.g., email, date).
4. 4.Define complex structures with grammar-based generation (e.g., JSON).
5. 5.Create reusable functions with @guidance decorator for reusable patterns.
6. 6.Integrate with tools via stateful functions (e.g., ReAct agents).