hanzoai/guard

Guard

LLM I/O sanitization and policy enforcement

Stop prompt injection, PII leaks, and unsafe outputs before they reach your users. Guard sits between your application and any LLM provider, enforcing policy on every request and response.

Defense in depth for LLM apps

A layered set of detectors and policies that ship with sane defaults and extend to your domain.

Prompt injection

Detect direct and indirect injection attempts in user input, retrieved documents, and tool outputs.

PII redaction

Mask names, emails, phone numbers, card data, and custom entities before they hit the model.

Output filtering

Block toxicity, jailbreaks, secret leakage, and topics your policy disallows on the response side.

Policy as code

Declarative rules per route, per role, per model. Version-controlled, reviewed, and auditable.

Full audit trail

Every decision logged with the rule that fired, the matched span, and the action taken.

Drop-in proxy

Point your OpenAI client at Guard. Same API, same SDKs. No application code changes required.

Get started with Guard

Open source

License: Apache-2.0hanzoai/guard

Get Guard

AI guardrails