Deepgram's Voice Agent API is a single-endpoint conversational AI platform that combines automatic speech recognition (Nova-3 model), LLM orchestration, and text-to-speech (Aura voices) into one real-time pipeline. Developers avoid stitching together separate STT, LLM, and TTS services. It achieves sub-300ms latency, includes built-in barge-in detection and turn-taking prediction, and supports bring-your-own-LLM or bring-your-own-TTS configurations. Primarily used for building customer-facing phone and web voice agents. Key features: - Single API unifying speech recognition, LLM orchestration, and voice synthesis - Sub-300ms end-to-end latency with 99.9% uptime SLA - Built-in barge-in detection, turn-taking prediction, and mid-session control - Function calling support for triggering backend actions during live conversations - Bring-your-own-LLM or bring-your-own-TTS while retaining Deepgram orchestration - HIPAA, GDPR, and regional data residency compliance; self-hosted deployment option
Voice Agent API: $4.50/hour flat rate. Pay-as-you-go after free $200 credit. Growth plans from $4,000/year with discounted rates. Underlying STT (Nova-2): ~$0.0043/minute; TTS (Aura): $0.0150/1,000 characters.