Skip to content

Guardrails

LLM Frameworks

An open-source toolkit for adding programmable guardrails to LLM-based conversational applications

Python Latest v0.22.0 · 12d ago Security brief →

Features

  • Add customizable guardrails to control LLM output behavior
  • Protect against jailbreaks, prompt injections, and other LLM vulnerabilities
  • Enable safe integration of models with external tools and services
  • Steer conversations along predefined paths for consistent interaction

Recent releases

View all 10 releases →
Config change
v0.22.0 Breaking risk
Auth

Usage reporting, LangChain optional, IORails enhancements

v0.21.0 Breaking risk
Breaking changes
  • Removed stream_usage from streaming metadata capture
Notable features
  • IORails Input/Output rail engine with parallel execution
  • check_async method for standalone rail validation
  • OpenAI-compatible server with v1/models endpoint
v0.20.0 Breaking risk
Breaking changes
  • Removed streaming field from configuration
Notable features
  • Nemotron reasoning-capable content safety models
  • GLiNER for open-source PII detection
  • Multilingual refusal messages in content safety rails
v0.19.0 Breaking risk
Notable features
  • LangChain 1.x compatibility
  • Content blocks API support for reasoning traces and tool calls
v0.18.0 Breaking risk
Breaking changes
  • Reasoning traces extracted to separate field instead of prepended to message
  • stream_async now raises error if output rail streaming disabled
  • Replaced reasoning trace extraction with LangChain additional_kwargs
Notable features
  • BotThinking events for processing reasoning traces
  • Embedding providers: Azure OpenAI, Cohere, Google
  • In-memory LFU caches for content-safety, topic-control, jailbreak models

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

About

Stars
6,344
Forks
710
Languages
Python Shell Dockerfile

Install & Platforms

Install via
pip
Platforms
linux macos windows

Beta — feedback welcome: [email protected]