Source: Guardrails AI library and docs

README and concept docs for the open-source Guardrails AI Python library: Guard objects, Hub validators, on_fail actions and the OpenAI-compatible server.

Contents

Summary

Guardrails AI is a Python framework that wraps LLM calls in a Guard composed of validators. Each validator checks one property (regex, toxicity, PII, competitor mentions, schema) and declares what to do on failure. It also generates structured output from a Pydantic model.

Key claims

  • Guard().use(Validator, ..., on_fail=...) builds a guard; guard.validate(text) checks text; guards can run on inputs and outputs.
  • Validators come from the Guardrails Hub. A 2026-07-06 notice says validators are moving to plain PyPI packages and hosted remote inference ends 2026-08-25.
  • Guard.for_pydantic(output_class=...) gets structured data, via function calling where supported or by putting the schema in the prompt otherwise.
  • On-fail actions: NOOP (default; pass through, log), EXCEPTION (raise), REASK (re-prompt the LLM with the validator’s error message), FIX (programmatic correction), FIX_REASK (fix, re-validate, reask if still failing), FILTER (structured output only: drop the failing field), REFRAIN (return nothing). Custom handlers are allowed.
  • guardrails start --config=... serves guards as a REST service with an OpenAI-compatible endpoint per guard, so clients only change base_url.

Feeds

Guardrail tooling

Suggest changes →