Realtime / Live API and Voice Agents Through the Gateway

A bidirectional, low-latency streaming API over WebSocket where a model receives and produces audio directly — speech-to-speech — instead of going through separate transcription and text-to-speech steps. OpenAI calls it the Realtime API; Google calls it the Live API.

The Realtime API is a single stream with lower latency and provider-managed turn-taking, but costs more per minute. The pipeline uses three separate services, which is cheaper and more tunable but adds latency. The gateway supports both.

It’s priced in audio tokens. OpenAI Realtime is roughly $0.30 per minute and Gemini Live roughly $0.03 per minute at current rates, and both re-bill context each turn. Verify current figures on the providers’ pricing pages.

OpenAI, Google Gemini, Google Vertex AI, and Azure AI Foundry, each via its native SDK pointed at the gateway’s WebSocket URL.

Yes. Because the client connects to the gateway rather than a provider directly, switching from OpenAI Realtime to Gemini Live is a configuration change, and you can split traffic to compare them.

Yes. TrueFoundry’s data plane runs in your VPC or on-prem, so DSPy’s prompts and optimization data stay in your environment.