Define your rules
Choose what you want to detect and control: prompt injection, jailbreaks, sensitive data, abuse, or disallowed content. Use QDivZero's built-in rules or create policies tailored to your application.
LLM Firewall in QDivZero
Protect every request before it reaches your model.
Block prompt injection, jailbreaks, disallowed content, and sensitive-data exposure before inference. QDivZero uses LLM Firewall to analyze every request, apply your security policies, and decide what can reach your models through an OpenAI-compatible API.
01 / Security before inference
Configure your rules. QDivZero applies them.
Configure how you want to protect your AI applications, and QDivZero applies those rules automatically to every request. Start with ready-made protections or create your own policies to adapt Firewall to each application, model, or use case.
Choose what you want to detect and control: prompt injection, jailbreaks, sensitive data, abuse, or disallowed content. Use QDivZero's built-in rules or create policies tailored to your application.
Every request passes through LLM Firewall before inference. QDivZero evaluates the content against active policies to identify risks or violations before sending it to the model.
Decide what should happen when a rule matches. The request can continue, be blocked, audited, or modified according to the policy you configured.
02 / Security rules
Start with a ready-made foundation and adapt it to your application.
QDivZero includes ready-to-use rules that protect LLM applications from common risks and lets you create your own controls when you need something more specific. Combine both options to build LLM security policies tailored to your models, users, and use cases.
Enable ready-made protections that detect prompt injection, jailbreaks, sensitive data, abuse, and disallowed content without developing every security control from scratch.
Create your own rules to control situations specific to your application. Define what you want to detect and how LLM Firewall should respond when a request meets those conditions.
Group built-in and custom rules into one policy to keep a common protection baseline and add controls specific to each application, model, or environment.
03 / Connected protection
When an AI application is in production, you cannot always trust what reaches your models. LLM Firewall lets you define what can pass through and apply the same security policies before every inference.
Detect and block prompt injection and jailbreaks before they can alter the instructions or behavior of your AI models.
Detect personal data, sensitive information, and PII in prompts, and apply your policies before sending them to the model.
Create guardrails and LLM security rules tailored to your application, users, and specific requirements.
Define which requests can reach your models and which must be blocked according to your AI security policies.
Add a security layer for LLM applications while keeping an OpenAI-compatible API and your existing inference workflow.
Use an evaluator model to analyze every request in context and apply rules that require semantic evaluation.
04 / FAQ
Everything you need to know to protect LLM applications, configure guardrails, and control which requests reach your models.
An LLM Firewall is a security layer for applications that use language models. It analyzes requests and applies policies to detect risks such as prompt injection, jailbreaks, sensitive data, or disallowed content before they reach the model.
Prompt injection protection analyzes inputs to detect attempts to manipulate the model's instructions or behavior. QDivZero LLM Firewall applies these protections automatically before inference.
LLM guardrails are controls that define what an AI application can do or process. An LLM Firewall centralizes those guardrails and other LLM security policies in a single protection layer.
Yes. An LLM Firewall can use rules to identify jailbreaks and attempts to bypass an application's restrictions. In QDivZero, you can combine built-in protections with rules tailored to your use case.
You can detect PII, personal data, and sensitive information before sending it to the model, then define what should happen when it appears in a request, such as blocking it or applying another security policy.
Yes. QDivZero lets you create custom rules and guardrails for situations specific to your application and combine them with LLM Firewall's built-in protections.
You can add an LLM security layer to the request flow while keeping an OpenAI-compatible integration. This lets you apply policies and guardrails without rebuilding your application's core logic.
Yes. Security policies are applied independently of the target model, so you can maintain a consistent protection layer when working with different LLMs and AI models.
It depends on the configured policies. You can define rules for prompt injection, jailbreaks, sensitive data, disallowed content, abuse, and other behaviors specific to your application.
When an LLM application is exposed to users, agents, or other systems, you cannot control every input it receives. An AI Firewall or LLM Firewall applies security policies consistently before requests reach your models.
Add LLM Firewall to your AI applications to detect prompt injection, jailbreaks, sensitive data, and disallowed content before inference. Configure your own security policies and control which requests can reach your models through an OpenAI-compatible integration.
Access the platform