Grodmir — self-hosted AI gateway

Grodmir is a self-hosted AI gateway for connected tools that need a shared data-policy boundary before model requests leave customer infrastructure. It inspects requests in the customer’s data plane, applies signed policies and blocks on detector errors; channels that bypass it remain unprotected.

Grodmir is an AI gateway for data protection and AI usage governance in organisations.

When is an AI gateway useful?

Grodmir is an AI gateway for connected tools that need a shared data-policy boundary before model requests leave customer infrastructure. Evaluate the channels you can route through it and test detection on your own data.

Link to this answer

When does it block or forward?

In enforce mode, detected data is transformed or blocked according to signed policy; detector errors, timeouts and missing valid policies fail closed. Observe mode forwards content unchanged and applies no policy blocks; authentication, admission and routing checks still apply.

Link to this answer

How can you evaluate the boundary?

Start with observe mode and synthetic requests, review metadata reports, then test enforce mode and failures. Use the product page for the data-plane and control-plane architecture.

Link to this answer

When should a software house evaluate Grodmir?

Grodmir is a candidate for IT companies and software houses that can deploy a gateway in their own infrastructure, route supported applications through it and maintain signed policies for organisations, end clients and projects. Evaluate it with synthetic requests and measure detection on your own data before enabling enforcement.

Link to this answer

Is Grodmir a self-hosted AI gateway?

Yes. The Grodmir inspection gateway runs in customer infrastructure, while configured model destinations receive the content permitted by policy. A portal account does not install the gateway. Self-hosting does not prevent direct connections from bypassing it, and observe mode forwards content unchanged.

Link to this answer

How does an AI gateway differ from an MCP proxy?

An AI gateway inspects model requests between an application and its configured model provider. An MCP proxy inspects a different connection: the arguments an agent sends to an MCP tool and the results returned to the agent. Grodmir supports both paths through the customer gateway, but each connection must be configured separately. Changing the model endpoint does not put directly connected MCP servers behind the proxy.

Link to this answer

Where do model-provider API keys belong?

Connected applications authenticate to Grodmir with virtual gateway keys; model-provider credentials stay in the gateway. This separates application access from upstream provider credentials. A virtual key only controls requests made through the gateway: an application using a separate provider key and a direct endpoint can bypass it, so network and device controls remain part of rollout.

Link to this answer

How can you verify this behaviour?

Content updated:

Protection covers connected-channel traffic only; detection can miss data and raise false alarms. Grodmir — AI gateway protection limits

Categories: AI gateway · Prompt DLP · MCP proxy · AI usage governance

Organisation portal