Demo library GA 4.13.4
← All demos

Gravitee AI Gateway: Secure, Govern, and Control AI Spend

One gateway can hold per-team AI budgets, step requests down to cheaper models and block overspend, with every call visible in the console

Pre-release build Customer safe

Pre-release, cleared for customers: Cleared by Sam on 29 Oct 2026. Recorded on the nightly image and shows pre-release behaviour (budget-based routing, Semantic Assessment Labs policy)

Made for Zilch

Say while it plays

  1. One gateway endpoint sits in front of Bedrock in the EU, Anthropic direct and a self-hosted model: callers ask for a tier, and the gateway can override that once their budget is spent.
  2. Policy Studio chains the controls left to right on the request and response flow: PII filtering, semantic caching, a cost limit, budget-based routing and a classifier that sends simple calls to a cheap model.
  3. The logs show each call's payload, tokens, model and remaining budget, and a call over budget is blocked at the gateway with a cost-budget-exceeded error and never reaches the model.

Don't claim

  • Don't present Meridian Pay as a customer: the recording says it is a fictional company.
  • Don't claim the traffic was live: the recording says it was fired through before the recording.
  • Don't claim this is the full feature set: the recording calls it a quick overview of a small subset.

Chapters

  • Introduction and problem statement
  • Policy Studio: routing, tiers, and budgets
  • Logging and observability dashboards
  • Advanced features: MCP proxies, tracing, export
  • Conclusion and next steps

Related demos