tools
Guardrails AI
Open-source Python framework and Hub for validating LLM inputs and outputs against safety, quality, and formatting risks.
Guardrails AI lets developers add input and output validation to LLM applications. Its Hub provides validators for PII, hallucinations, jailbreaks, prompt injection, toxicity, formatting, SQL, and other risks, which can be combined into Guards.
It is used for chatbots, RAG systems, structured data extraction, code generation, summarization, and agent workflows. The framework is open source, but hosted remote inference was discontinued on August 25, 2026; validators now run locally or on infrastructure managed by the user.
Features
- Validate LLM inputs and outputs with configurable Guards
- Browse and install validators from the Guardrails Hub
- Detect PII, prompt injection, jailbreaks, toxicity, and hallucinations
- Generate structured data from LLMs
- Apply corrective actions including reask, fix, filter, or exception
- Run validators locally or on a user-managed inference endpoint
- Apache 2.0 open-source framework
Use cases
- Validate chatbot responses before returning them to users
- Check RAG answers against source documents
- Detect PII and secrets in generated text
- Constrain text-to-SQL and code-generation outputs
- Enforce formatting, topic, tone, and policy rules
- Test agent workflows for unsafe or unexpected behavior