Hi LiveKit Team! I opened a feature request around exposing LiveKit Inference cost/pricing programmatically: Expose LiveKit Inference cost/pricing through AgentSessionUsage or a pricing API · Issue #7420 · livekit/agents · GitHub
AgentSessionUsage class already gives us normalized usage per provider/model, which is great for observability. The gap I’m running into is cost.
For BYOK/provider plugins, I understand LiveKit can’t know the customer’s actual rate because that can vary by provider plan/contract. ( if this could be possible? )
But for LiveKit Inference, LiveKit is already metering and billing the model usage, so it seems like there should be an authoritative way to get either:
the cost directly from AgentSessionUsage, or
a machine-readable pricing API/catalog that maps model + usage unit → LiveKit Inference rate.
I maintain an open-source pricing catalog for voice models here: GitHub - mahimailabs/voice-prices: Open price database for voice AI APIs - STT, LLM, TTS, S2S, VAD · GitHub
If LiveKit exposes the pricing source, I’m happy to consume it there instead of maintaining a mirrored copy.
Curious which direction makes more sense from the LiveKit side.
Thanks!