LLM Firewall
Protect model traffic
Evaluate supported chat requests against reusable policies before they reach a model.
How protection works
A firewall combines rules, a judge model and an enforcement mode. The client sends the firewall slug with the request; QDivZero evaluates the input before inference.
Choose the mode
Audit records decisions without interrupting traffic. Block rejects matching requests. Allow limits traffic to requests that satisfy the configured policy.
Keep policies reusable
Maintain rules independently and combine them into different firewalls for products, environments or risk profiles.