Guardrails AI is an open-source (Apache 2.0) reliability platform that intercepts LLM outputs before they reach users, running validation logic to catch hallucinations, PII leakage, policy violations, and unsafe content. It offers a Guardrails Hub with 70+ prebuilt validators and supports custom validators. The platform also covers synthetic data generation and pre-production evaluation. Enterprise customers include Robinhood. A managed cloud service, Guardrails Pro, adds hosted validation, observability dashboards, and dedicated support on top of the free self-hosted core. Key features: - Runtime output validation and correction before responses are delivered - Guardrails Hub with 70+ prebuilt validators (PII, jailbreaks, factuality, code exploits, brand risk) - Synthetic data generation for fine-tuning and prompt testing - Dynamic evaluation datasets for pre-production edge-case testing - Near-zero latency impact (typically 10-50ms overhead per response) - Self-hosted open-source core plus managed Guardrails Pro cloud tier
Open-source core is free (Apache 2.0). Guardrails Pro (managed cloud) pricing not publicly listed; contact sales.