FAQ

How much latency does it add?

The deterministic path (policy, regex, loop breaker) is under 2 ms. If the AI Guard is enabled, semantic inspection adds roughly 50–100 ms (TypeSafe AI Jev). With the AI Guard off there is no extra model latency.

Is my data sent to AvertGate?

Not by default. Tool calls, results, file contents and queries are processed in your own network. Only redacted audit events go to the control plane. If you enable the AI Guard with a cloud provider, the content it inspects is sent to that provider. By choosing a local model for the AI Guard you can keep that path entirely inside your network.

What happens if AvertGate goes down?

Because the gateway is inline, tool calls don't go through while it is down; this is deliberate fail-closed behaviour. For high availability, run several instances and use cluster mode.

Which models are supported?

It is independent of the LLM your agent uses; AvertGate works at the tool layer. For the AI Guard you can use TypeSafe AI Jev, Ollama, OpenAI-compatible endpoints, Anthropic or a custom HTTP endpoint.

How should I try it first?

Start in shadow mode, review the SHADOW_BLOCK records in the dashboard, tune the policy and then switch to enforcement.

Pricing?

An annual enterprise contract. Contact us for details.

What is available for a security review?

See the Security & data handling page. For a security questionnaire or an architecture review, get in touch.