Generation

Patterns - Generation is optional and separate from verification. - Interface: Generator protocol (generate(prompt, context_pack, temperature) -> dict). - Reference: LLMGenerator uses an LLMProvider to produce a JSON answer, prompting with a facts-first context pack and “answer only if supported” instructions.

Customize - Override LLMGenerator.build_prompt to change formatting or add domain instructions. - Override LLMGenerator.build_schema to add fields (citations, confidence). - Implement your own Generator to add streaming, guardrails, or different output formats.

Example

from contextguard import LLMGenerator, OpenAIProvider

llm = OpenAIProvider(model="gpt-4o-mini")
gen = LLMGenerator(llm)
result = gen.generate("Summarize ACME revenue.", context_pack)