guardrails

Overview ¶

This example demonstrates output validation guardrails:

• Basic guardrails: LengthGuardrail, ContainsGuardrail, BannedWordsGuardrail
• Advanced guardrails: LLMGuardrail (LLM-as-a-judge), HallucinationGuardrail (fact-check), ContextGuardrail interface
• Chain (multi-guardrail orchestration with collect-all / stop-on-first / max-errors)
• Structured error handling: ValidationError / ValidationFailure with errors.As

Prerequisites:

• An LLM API key (OpenAI, Ollama, or any supported provider)

Usage (OpenAI):

OPENAI_API_KEY=sk-... go run main.go

Usage (Ollama):

OLLAMA_BASE_URL=http://localhost:11434 OLLAMA_MODEL=gemma3:4b go run main.go