guardrails overview

guardrails_overview.md

Tool/Library	Guardrail Definition	Guardrail Application	Supported Types	Violation Handling	Open Source?	Code Example
AWS Bedrock Guardrails	Rules, constraints, policies	Training, inference	Safety, security, fairness, privacy	Exceptions, warnings, output modification	No	bedrock_client.create_guardrail(name='my_guardrail', ...)
Nvidia Nemo Guardrails	Colang flows	Training, inference	Input, dialog, output, retrieval, execution	Reject input, alter input, execute actions, reject output, alter output	Yes	config = RailsConfig.from_path("path/to/config")
Guardrails Python library	RAIL specifications	Inference	Semantic validation, type enforcement	Corrective actions, re-prompting	Yes	guard = gd.Guard.from_rail_string(rail_str)
LangChain	Integrates with Guardrails Python library	Inference	Content moderation, data privacy, bias mitigation	Filter, fix, exception	Yes	chain = prompt
Google Checks API	Policies	Inference	Dangerous content, PII, harassment, etc.	Logging, blocking	No	service.aisafety().classifyContent(body={'input': ...})
Arize	Dataset embeddings, LLM evaluators	Inference	Jailbreak detection, PII detection, toxicity detection	Default response, re-prompting, exception	Partially	guard = Guard().use(ArizeDatasetEmbeddings, ...)
Giskard	Colang rules	Inference	Vulnerability detection, safety, security	Generate Colang rules for mitigation	Yes	scan_report.generate_rails("config/generated_rails.co")
Galileo	Guardrail metrics	Inference	Quality, correctness, safety	Alerts, monitoring	No	Not available