Voximplant AI is a serverless runtime for Voice AI pipelines that connects real-time agent/LLM systems and speech engines to PSTN / SIP / WebRTC / mobile / WhatsApp calling, with code-driven orchestration and provider flexibility. See Voximplant AI and the docs Voice AI connectors section.

Supported vendors (direct agent / real-time LLM connectors)

Native/direct connectivity is positioned for:

• OpenAI (Realtime / agent-style integrations) - Docs: OpenAI
• Google Gemini (Live) - Docs: Google
• Deepgram Voice Agent - Docs: Deepgram
• ElevenLabs Agents / Conversational AI - Docs: ElevenLabs
• Ultravox (WebSocket API) - Docs: Ultravox
• Cartesia Line Agents - Docs: Cartesia Line Agents
• xAI (Grok Voice Agent) - Docs: xAI

Voximplant AI also explicitly supports connecting to another WebSocket interface (for other real-time AI systems) in addition to the vendors above.

Supported vendors (speech engines: STT / TTS)

Voximplant’s platform speech layer (STT/TTS) includes built-in providers such as:

• Speech-to-Text (STT): Google Speech Cloud, Microsoft Azure STT, Amazon Transcribe, Yandex Speech Cloud
• Text-to-Speech (TTS): Google Speech Cloud, Amazon Polly, Yandex Speech Cloud, Microsoft Azure TTS, Tinkoff VoiceKit

For realtime / streaming TTS used in Voice AI scenarios, Voximplant also provides native VoxEngine modules and guides for:

• Cartesia Realtime TTS - Guide: Realtime TTS and API refs
• Inworld Realtime TTS - Guide: Realtime TTS
• ElevenLabs Streaming / realtime TTS - Guide: ElevenLabs TTS and API refs

Pipeline options (architectures you can run)

• Speech-to-speech: real-time audio in and real-time audio out (agent API handles full duplex loop)
• Speech -> LLM -> TTS: stream audio directly into a speech LLM and use a different TTS for output
• STT -> LLM -> TTS: stream audio to STT, pass text to an LLM/toolchain, synthesize response audio
• Hybrid: combine a real-time agent API for turn-taking with separate best-of-breed STT/TTS components (mix and match)

Orchestration primitives (what you control)

• Mix and match providers: swap STT/TTS/LLM vendors without changing your telephony integration
• Parallel model execution: run multiple speech/LLM components in parallel when useful (for example, intent extraction + generation)
• Failover paths: fall back to alternate speech/LLM providers when a step errors or times out
• Wideband audio: higher fidelity audio path for improved user experience and model comprehension
• Deep SIP support: SIP trunking + registration interop so agents can operate inside PBX/SBC/carrier environments
• Channel portability: reuse the same AI pipeline across PSTN numbers, SIP, WebRTC, mobile SDKs, and WhatsApp calling

Real-time media integration (streaming)

• WebSocket-based media streaming for connecting calls to real-time AI systems and custom pipelines (audio + metadata/control messages on the same channel)
• Media gateway abstraction: avoid building/operating custom streaming gateways when using native connectors/modules

Connectivity and endpoints

• PSTN calling (inbound/outbound) via phone numbers and programmable call handling
• Phone numbers API: automated procurement in 60+ countries (availability varies by country)
• SIP calling and trunking: connect carriers / PBXs / SBCs using SIP interop (including registration-based scenarios)
• WebRTC calling via web/mobile SDKs (VoIP calling in apps and browsers)
• WhatsApp calling: inbound/outbound voice calls via WhatsApp Business API integration

Serverless call control (VoxEngine)

• JavaScript call logic (no XML) for real-time call routing and application workflows
• Per-call-leg signaling/media control - granular control over each leg independently

Conferencing and bridging

• Single conferencing API for voice/video; mix PSTN, SIP, WebRTC, and native mobile endpoints
• Conferences up to 50 participants

Recording, transcription, and speech processing

• Call recording via call.record() in scenarios (supports stereo and additional options)
• Call transcription via record(transcribe=true) and retrieval via GetCallHistory (transcription delivered asynchronously)
• Speaker/channel labeling in transcripts (for example, “Left”/“Right” labeling pattern described in docs)

Speech-to-Text (ASR) modes and features

• Phrase-hint mode (best for constrained dialogs / IVRs) and Freeform mode (open transcription)
• Multiple ASR engines (for example, Google, Amazon, Microsoft, Yandex, T-bank) with selectable profiles
• Intermediate results support (provider-dependent) for faster partial recognition
• Google Speech v1p1beta1 feature passthrough (for example, word time offsets, punctuation, diarization config)

Answering machine / voicemail / beep detection

• AMD module for voicemail/answering machine detection in scenarios
• Beep detection with specified frequency lists and timeouts (scenario-level control)
• AMD event/callback model available in VoxEngine references

Automated outbound calling (call lists + dialing logic)

• Call Lists: upload a CSV call list and process it with VoxEngine scenarios (campaign-style calling)
• Management API CallLists: programmatic call-list upload/append with delimiter support
• Predictive Dialing System (PDS): uses agent/load statistics and call-list progression to place calls and connect answered calls to agents
• Predictive and progressive dialing modes with tunable parameters (for example, allowed failed call percentage)

Cloud IDE and debugging

• Cloud IDE + debugger in the control panel:
• Code verification
• Autocompletion
• Diff highlighting
• Built-in troubleshooting workflow

** Local IDE continuous integration **

• CLI tool for CI/CD automation so you can use your own IDE
• Type library for local development with autocompletion and type checking

SDKs and client libraries

• SDKs: iOS, Android, Web, React Native, Flutter, Unity
• API clients: curl, Node.js, Python, PHP, Go, .NET, Java

Management API (HTTP)

• Control accounts/services programmatically (examples from docs include managing phone numbers, messaging, billing, logs, records, and user access)

• Media Streams: integrate live audio streams into calls via WebSockets for real-time transcription/analysis and AI integrations
• WebSocket programming model in VoxEngine:
• Create connections via VoxEngine.createWebSocket(...)
• Stream audio using WebSocket.sendMediaTo(...)
• Recommended audio chunk duration: ~20ms

• Serverless runtime: no infrastructure to manage for call logic
• CI/CD-friendly deployment path: CI tool for automated pipelines
• Global footprint: datacenters in 14 countries
• Status page for live and historical uptime of subcomponents