[FEATURE]: Add Model-Agnostic Guardrails Documentation Using LiteLLM and Hooks #1088
Description
Problem Statement
- Only Bedrock guardrails are documented
- No examples for LiteLLM-based providers (OpenAI, Anthropic, Nebius, etc.)
- Hooks are provider-specific
- Missing reusable, cross-provider guardrails pattern
Proposed Solution
Add a new section “Model-Agnostic Guardrails” under the Guardrails documentation with:
- Basic Implementation: Hook-based guardrails using BeforeInvocationEvent and AfterInvocationEvent
- Multi-Provider Examples: Show the same guardrails working with OpenAI, Anthropic, and Nebius via LiteLLM
- Advanced Patterns: Use a dedicated LiteLLM model for content safety evaluation
- Best Practices: Performance and reliability guidance
Use Case
People who're using Open Source Models via LIteLLM can also use Guardrails.
Alternatives Solutions
No response
Additional Context
I've implemented one similar use case here: AWS Strands Litellm Guardrails