Demo library GA 4.13.4
← All demos

Per-model AI cost without instrumenting the caller

Per-model cost and token usage are visible without instrumenting the caller

Recorded on 4.12 Customer safe

Say while it plays

  1. Every AI workload, LLM proxies, MCP proxies and A2A agents, shows up in one catalogue, and if it isn't on that page the gateway isn't routing to it, so shadow AI has nowhere to hide.
  2. The gateway reports tokens in, tokens out and per-model spend natively, which is the number finance keeps asking for and nobody can usually produce at a strategic level.
  3. You can drop into individual request traces to see which model served a call and what it cost, all from the same screen with no second dashboard or side channel.

Don't claim

  • Don't claim spend optimisation or advanced model-routing recommendations exist today: the recording says those are coming soon on the roadmap, not shipped.

Chapters

  • Why observability and cost matter
  • Unified catalog and gateway
  • Proxy routing and shadow‑AI prevention
  • FinOps reporting and upcoming optimisation

Related demos