Demo of Gravitee Gateway with Hotel Concierge Agent – Live Databricks Backend and Policy Controls

In this video I walk through a live demo that extends the hotel‑concierge example I showed at APIs Days Munich. I kept the original architecture – a back‑end API that reads and writes to a live Databricks data store – but added the Gravity Gateway so I can toggle the gateway on and off while the same traffic still reaches the same back‑end agent. The LLM routing is configured to try Grok first and fall back to Gemini Flash. User logging, authentication, and access‑management stay the same.

I built a new thin UI hosted at hotels.project.cc. From the UI you can launch the concierge agent and choose whether the request goes directly to the back‑end or through the Gravity Gateway. I scripted nine different scenarios that demonstrate typical actions (booking a stay, cancelling, querying user data) and you can also add your own.

When the gateway is OFF the agent can book a room, update a booking, and the change appears instantly in the Databricks instance. I then show what happens when the same actions are sent through the gateway: permissions are checked, policies are enforced, and the system can block or require human approval. For example, a “delete my booking” request triggers an over‑reaching‑permissions policy and the gateway returns a refusal. A “leak personal information” scenario shows the policy denying the request. I also illustrate token‑bomb protection – feeding a 2,000‑token transcript causes the gateway to reject the prompt before it reaches the model.

A human‑in‑the‑loop step is demonstrated: the gateway flags a risky request, presents an approval dialog, and I either approve or deny it. If denied, the agent apologises; if approved, the action proceeds and an OpenTelemetry log is written to the database, confirming the operation.

I then compare the results with the gateway OFF: the same prompt returns data for all database members, highlighting the importance of policy enforcement. The demo is meant as a visual guide for teams starting to manage AI agents and shows the power of the Gravity Gateway for security, cost‑control, and observability. I mention that the code is available for anyone who wants to dig deeper, and I wrap up by noting that this video mirrors the workshop I gave at API Studies in Munich and invites viewers to reach out with questions.