Full-duplex with interruption
The agent listens while it speaks. Customers can interrupt mid-sentence; the agent yields, pivots, and resumes context-aware.
Brand voice cloning
Train on 50 reference calls to produce a custom voice. Train on 200 for cinematic, emotion-tuned brand voice.
Sub-500ms end-to-end latency
STT, LLM reasoning, and TTS synthesis stitched into a single low-latency pipeline. Conversation feels human.
Multi-channel routing
One agent definition runs on PSTN, SIP trunk, browser SDK, and in-app voice. Stateful across channels via session ID.
Compliance-aware redaction
PCI, PII, and health identifiers detected and redacted from transcripts in real-time. Audit trail per conversation.
Hot-swap LLM backends
Switch between GPT, Claude, or self-hosted models per workflow. A/B test variants without redeploying.
Cross-channel context
Conversation history persists across web, app, WhatsApp, and SMS. Customer never repeats themselves regardless of channel switch.
Visual UI generation
The agent renders cards, carousels, forms, and action buttons inline. Customer interacts without leaving the chat thread.
Async + sync modes
Handles real-time chat AND multi-day async threads. WhatsApp customers get response when convenient; web visitors get instant reply.
Streaming responses
Token-by-token streaming reduces perceived latency. Customer sees progress while the agent thinks.
Multi-language support
Native English (en-NZ, en-AU, en-UK, en-US) plus French, Spanish, German, Italian, Dutch via locale switch. More languages on request.
Audit-grade logging
Every message, intent classification, tool call, and handoff event logged with timestamps. Exports to your compliance system.
Real conversations. Any vertical.
Pick a vertical. Pick a use-case. The agent runs the conversation — and the integrations light up as it does.
One voice runtime. Three quality tiers.
The same full-duplex stack — STT, LLM orchestration, TTS, interruption handling — at three voice fidelities. Pick by use case and budget; mix and match across campaigns.
Voice or chat? Both, usually.
Same platform, different surfaces. Pick by where the conversation naturally lives — high-trust live interactions belong on voice, transactional and document-heavy work belongs in chat.
- PRIMARY SURFACEVOICESIPCHATWeb, WhatsApp, SMS, RCS, email
- TURN-TAKINGVOICEFull-duplex (interruptible)CHATTurn-based (typing indicators, async OK)
- END-TO-END LATENCYVOICESub-500ms (Ultra Premium / Premium tiers)CHATTypically 1–3 seconds
- IDENTITY / BRANDVOICECloned brand voice (50 reference calls)CHATBrand-tuned tone + house style guide
- VISUAL UIVOICECompanion SMS / email artefactsCHATIn-chat cards, forms, carousels (generated)
- BEST FORVOICEHigh-trust live interactions: collections, FNOL, healthcareCHATAsync, transactional, document-heavy: order status, returns
- OPERATES 24×7VOICEYesCHATYes
- HAND-OFF TO HUMANVOICESentiment- or rule-triggered queue transferCHATSame — escalation routes to your existing helpdesk
- COMPLIANCE HOOKSVOICECall recording, PCI redaction, audit trailCHATConversation log, redaction, audit trail
Frequently asked questions
Pricing, latency, residency, integrations — answered. Reach out to sales for anything not covered here.
Same underlying platform; different surfaces. Voice runs over PSTN, SIP, or browser SDK with brand voice cloning and full-duplex turn-taking. Chat runs over web, WhatsApp, SMS, RCS or email with visual UI generation and cross-channel context.
A 14-day pilot is the standard entry point, with about 30 days to production for a single scenario. Voice cloning needs around 50 reference calls (200 for cinematic quality). Chat scenarios ship faster because there is no voice training step.
Sub-500 milliseconds across STT, LLM and TTS combined, on the Ultra Premium and Premium voice tiers. Standard tier latency is higher and trades off cost for responsiveness.
Anthropic Claude and OpenAI GPT by default. Self-hosted open models are also supported. A/B testing across LLMs is built into the platform so a deployment can switch models without re-tuning the scenario.
The platform uses TLS 1.3 in transit, AES-256 at rest, zero-retention LLM routing, configurable regional data residency, and a per-conversation audit trail. A DPA and a security questionnaire are available on request. Formal third-party certifications are not currently claimed; the relevant programmes are on the roadmap and will be communicated as they land.
New Zealand by default. Australia, the United States, the United Kingdom and the European Union are available on request. Cross-region replication is opt-in only.
Voice is priced per conversation minute across three tiers (Standard, Premium, Ultra Premium). Chat is priced per resolved transaction. Pilots are fixed-fee. Final pricing is bespoke per deployment.
English variants — en-NZ, en-AU, en-UK, en-US — plus French, Spanish, German, Italian and Dutch through locale switching. Additional locales are available on request and added during onboarding.
Yes. Handoff is sentiment-triggered or rule-based, and routes to your existing contact-centre queue with full conversation context passed along.
Salesforce, HubSpot, Zendesk, SIP and PSTN telephony, WhatsApp Business, plus a REST API and webhooks for everything else. The platform is also MCP-compatible.