Build voice agents that scale
All the control of code. None of the infrastructure overhead.
Built for developers, from the first line.
Every primitive is exposed over a typed API and an SDK — so you stay in control of the experience.
Ultra low latency
Sub-500ms time-to-first-token keeps conversations feeling instantaneous, even over global carriers.
Natural conversation ability
Turn-taking, barge-in, and human-like voices so callers forget they're talking to an agent.
Live call tool & data access
Give agents tools at runtime — including MCP servers — to fetch data and take action mid-call.
Custom model support
Bring 200+ models and integrations, or wire in your own transcription, LLM, and TTS.
Telephony control
Custom SIP, warm transfers, voicemail detection, and DTMF — full control of the call path.
Enterprise grade uptime
Resilient infrastructure on the PCCW Global network with contractual SLAs.
Powerful for engineers, accessible for everyone
Configure in code or in the dashboard. The same primitives, your preferred workflow.
From silence to speech in minutes.
Deploy production-ready voice agents in minutes, then customize every interaction as you scale.
Give your agent context
Provide instructions and tools, and configure it as if you were building it yourself.
Pick your models
Choose from 200+ built-in integrations or connect your own transcription, LLM, and TTS.
Integrate your business data
Connect your authenticated internal APIs to fetch data and perform actions.
Tune and observe in production
Catch hallucinations before they go live with evals and an observability stack.
Deep control and flexibility for all builders.
Configure your agent your way. No lock-in, full control over every component.
Multilingual
Talk to agents in English, Spanish, Mandarin, or one of 100+ other supported languages.
Automated testing
Design test suites to identify hallucination risk before going to production.
Tool calling
Intelligently access data and perform actions across your applications.
Reporting & Analytics
Dashboards and exports to track performance, usage, and conversation quality.
Bring your own models
Bring your own API keys for transcription, LLM, or TTS — or plug in self-hosted models.
A/B experiments
Test variations of prompts, voices, and flows to continuously optimize performance.
Start a call in a few lines.
A typed SDK for every runtime. Create an assistant and place a call in seconds.
import { PCCWGVoice } from "@pccwg/voice-agents-server-sdk";
const client = new PCCWGVoice({ apiKey: process.env.PCCWG_API_KEY });
const call = await client.calls.create({
assistant: {
name: "Support Triage",
model: { provider: "openai", name: "gpt-4o" },
voice: "PCCW Aria",
firstMessage: "Thanks for calling PCCW Global. How can I help?",
tools: [{ type: "mcp", server: "https://api.internal.tools/mcp" }],
},
phoneNumber: "+85251234567",
});
console.log("Call started:", call.id);Customize every layer of the stack.
Compose the exact pipeline your use case needs, on infrastructure that scales with you.
Multi-Agent Orchestration
Chain or squad agents for complex workflows with intelligent handoffs and routing.
Modular Platform
Swap transcription, LLM, and TTS independently — no lock-in to any single provider.
Model Fallbacks
Configure automatic fallbacks so a provider outage never drops a call.
API-First Architecture
Everything the dashboard does is available over a typed, versioned REST API.
Build your first voice agent today.
Start free, then scale to millions of calls on the PCCW Global network.