MEETFIREWALL.

Protect every request before it reaches your model.

Firewall sits in front of your AI models to evaluate every request before inference begins. Apply production-ready guardrails, enforce custom policies, and decide whether requests should continue, be blocked, or be audited—all through a single OpenAI-compatible API.

Start with ComputeSee built-in rules

Control before inference.

AI applications shouldn't rely on prompts alone to stay safe. Firewall introduces a dedicated enforcement layer that evaluates every request before it reaches the model, giving teams a simple way to apply policies, protect sensitive data, and operate AI systems with confidence.

See how it works.

TURNPOLICIESINTO ACTION.

Decide what happens to every request.

Firewall evaluates each request against built-in and custom rules, then applies the configured action. Allow it, block it, or pass it forward with guardrail context.

Go to platform

CONTROL THEOUTCOME.

Enforce policy without forcing every request down the same path.

Firewall lets you choose how every policy is enforced. Continue matching requests with guardrail context when guidance is enough, or block them before they ever reach the model.

Allow with context

Guide the model.

When a request matches a rule, Firewall can pass it forward with guardrail context. The model still responds, but it receives the information needed to handle the request safely.

Firewall

Model

Block before inference

Block it early.

For stricter policies, Firewall returns an error before the target model is invoked. The request stops immediately and does not consume target-model compute.

Request

Model

Prompt injection

Detects instructions designed to override or manipulate model behavior.

Prompt inspection

system
Ignore previous
Override blocked

PRODUCTION-READYRULES.

Protect common AI risks from day one.

Firewall includes a curated set of production-ready policies that help prevent common risks such as prompt injection, sensitive data exfiltration, abuse, and unsafe content. Enable them instantly, keep them updated automatically.

YOUR POLICIES. YOURLOGIC.

Create custom rules in plain language.

Firewall evaluates every request against the policies you define and applies the selected action automatically. Use the same allow, block, and audit modes as built-in rules without building your own moderation pipeline.

Plain text
policy =

What Firewall includes

Built-in rules

Maintained coverage for common

AI safety, privacy, abuse,

and security risks.

Custom rules

Define policies in plain language.

Apply them through the same

Firewall enforcement path.

Allow mode

Continue matching requests

with guardrail context for

safer model responses.

Blocked mode

Return an error before target

model invocation, keeping

blocked requests out.

OpenAI-compatible

Attach Firewall to supported

chat completions without

changing your SDK flow.

Evaluator model

Choose the evaluator model

that reviews policies before

a request reaches its target.

Python
from openai import OpenAI

client = OpenAI(
    base_url="https://api.qdiv0.com/v1",
    api_key="your-api-key",
)

response = client.chat.completions.create(
    model="moonshotai/Kimi-K3",
    messages=[
        {"role": "user", "content": prompt},
    ],
    extra_body={{"firewall": "production-guardrails"}},
)

PROTECTIN ONELINE.

Add Firewall with a single request field.

Attach a Firewall policy to your request, and every prompt is evaluated before it reaches the model. No additional infrastructure, proxies, or SDKs required—just reference the policy and Firewall handles the rest.

View docs

Ship AI
with
confidence.