Guardrails AI

Guardrails AI is an open-source Python and JavaScript framework (Apache 2.0, roughly 6,600 GitHub stars) for validating and structuring LLM outputs. You declare what you want — JSON structure, content constraints — attach validators from the Guardrails Hub (dozens of pre-built checks from regex matching to toxicity, PII detection, competitor mentions, bias filtering, and hallucination detection via LLM judges), and Guardrails calls the model, checks the output, and automatically re-prompts when validation fails. It handles structured data extraction with Pydantic models and ships as a Python package, JS library, CLI, and Flask API server.

The project remains actively developed at v0.11.0 as of August 2026, with a 2026 roadmap covering token-level streaming validation, multimodal validators, and Guardrails Studio, a visual editor. In July 2026 the team announced validators are moving to standard PyPI packages with hosted remote inferencing discontinued — a simplification for self-hosters.

Limitation: the hosted remote inferencing cutoff (planned August 25, 2026) means teams relying on the cloud path must migrate to local validator packages, and validator quality varies — Hub validators are community-contributed, so safety-critical deployments should audit the specific validators they use rather than trusting the catalog wholesale.

Install with pip install guardrails-ai; validators are installed as standard PyPI packages from the Guardrails Hub.