UltraVoice Platform
Build real-time voice AI applications with enterprise-grade reliability. Stream bidirectional audio over WebSocket, plug in any LLM, ASR, or TTS provider, and deploy to phone lines in minutes.
AT A GLANCE
300 – 500 ms
End-to-end latency (ASR → LLM → TTS)
44 AI Models
25 LLM · 9 ASR · 10 TTS providers
Telephony
Twilio
What You Can Build
UltraVoice gives you a complete backend for voice AI — from WebSocket sessions for web apps to fully automated inbound and outbound phone calls.
USE CASES
Customer Support Bot
Answer calls, resolve tickets, escalate to humans — 24/7 with no hold times.
Voice Onboarding
Walk new users through setup steps using a natural two-way conversation.
Appointment Booking
Let callers schedule, reschedule or cancel via phone with calendar integration.
Outbound Campaigns
Trigger bulk outbound calls with dynamic context per contact.
Real-time Voice UI
Embed a voice chat widget in your web app using the WebSocket SDK.
AI IVR Replacement
Replace menu trees with a conversational agent that understands intent.
Key Features
PLATFORM FEATURES
Voice Agents
Create agents with a system prompt, welcome message, language, and voice. Each agent owns its provider config.
Knowledge Base
Inject workspace-level or per-session knowledge into every conversation. Update content without redeploying.
Custom Tools
Give agents REST API tools. The LLM decides when to call them — no extra prompt engineering needed.
Voice Test
Test your agent in the browser with a real microphone before connecting to a phone number.
Telephony
Receive inbound calls or trigger outbound calls with any of 6 SIP/PSTN providers via a single API.
Provider Registry
Swap LLM, ASR, or TTS providers per agent with no code changes. 44 models available today.
How the Pipeline Works
Every voice session runs the same real-time cascade:
PIPELINE STAGES
1 · Receive Audio
Raw PCM audio streams in over WebSocket or a telephony provider webhook.
2 · ASR (Speech → Text)
Real-time transcription via Deepgram, AssemblyAI, Google, Azure, or 5 others.
3 · LLM (Reasoning)
Streamed response from OpenAI, Anthropic, Groq, DeepSeek, Mistral, or 20 others.
4 · TTS (Text → Speech)
Audio synthesised and streamed back — ElevenLabs, Cartesia, Deepgram, and more.
Agent interruption detection, silence handling, and idle timeouts are managed automatically by the pipeline.
Supported Providers
AI PROVIDERS
LLM — 25 Providers
OpenAI, Anthropic, Google, Groq, DeepSeek, Mistral, Fireworks, Cerebras, Ollama + 16 more
ASR — 9 Providers
Deepgram, AssemblyAI, Google, Azure, Gladia, ElevenLabs, Cartesia, USF Mini, Custom
TTS — 10 Providers
ElevenLabs, Deepgram, Cartesia, OpenAI, Azure, Google, LMNT, Playht, USF Mini, Custom
Base URL
https://ultravoice.us.incAll API endpoints are relative to this base URL. Pass your JWT token as a Bearer token in the Authorization header.
Authorization: Bearer <your_access_token>Get Started
New to UltraVoice? Follow this path to get your first call working: