Real-time rendering
Running at ~25fps, with natural motion and micro-expressions - not looped video or basic mouth-dubbing.
Technology
Real-time conversation, engineered for the real world.
Zigg.ai combines a best-in-class real-time avatar engine with hardware and software built for live venues - and a cost model that doesn't bill you for an empty room.
How a conversation happens
The stack at a glance.
- Step 1
A visitor approaches.
Built-in proximity sensors detect a person within ~1.5 metres.
- Step 2
The session opens.
A live WebRTC connection starts in under five seconds.
- Step 3
They speak; it listens.
Speech recognition and turn-taking handle natural back-and-forth, including interruptions.
- Step 4
It thinks, grounded.
The response is generated from the character's approved knowledge base, within set boundaries.
- Step 5
It answers, face to face.
Anam's real-time avatar renders the reply with natural lip-sync and expression - sub-second.
- Step 6
It resets.
After ~15 seconds of inactivity the live session closes and the unit returns to its idle loop.
The face layer
The avatar engine: powered by Anam.
Zigg.ai's conversational avatars run on Anam and its real-time CARA model.
Sub-second responsiveness
~180ms avatar-model latency and sub-second end-to-end conversational response.
Multilingual
5+ languages at launch, scalable to 50+ with native voices.
Custom voices
Including permitted voice cloning for authenticity.
Bring-your-own-LLM or built-in
The "brain" is configurable; the persona is defined at runtime.
Benchmark-leading realism
CARA ranks top of independent 2025 avatar benchmarks.
The avatar is the face; the grounding, boundaries and hardware around it are what make it venue-ready.
No idle billing
Proximity-triggered cost control.
Real-time avatar sessions are billed by the second - for the whole time a session is live, whether or not anyone is talking. Left running, an avatar quietly bills against an empty gallery all day. That's the "idle-state billing leak."
Zigg.ai closes it.
Local idle loop.
Between visitors, the unit plays a local 3D loop at £0 in avatar fees.
Sensor activation.
Infrared proximity sensing wakes a live session only when a visitor is actually present (~1.5m).
Live handshake.
WebRTC connects in under 5 seconds - fast enough to feel instant.
Auto-timeout.
The session closes after ~15 seconds of inactivity, returning to the free idle loop.
The result: you pay for conversations, not for standby.
Safety by design
Grounded, fenced and brand-safe.
Grounded answers.
Characters draw on approved documents and verified sources.
Boundaries you set.
Figures stay on-script and within defined limits - no drifting off-topic or off-brand.
Consent-first likeness.
Real, living people are only recreated with explicit consent or rights clearance. We won't build avatars to deceive, impersonate or mislead.
Enterprise-grade foundations.
Built on Anam's SOC-2 / HIPAA-grade, encrypted infrastructure.
Live in days, not months
Deployment timelines.
| Scenario | Timeline |
|---|---|
| Existing unit + existing character | 24–72 hours |
| Existing unit + simple new character | 3–7 days |
| Existing unit + custom voice / executive twin | 1–2 weeks |
| No hardware in stock | + ~3 weeks |
| Full bespoke project | 4–8 weeks |
A typical custom build runs: Weeks 1–2 research and define boundaries → Weeks 3–4 voice and content setup → Weeks 5–8 hardware install and venue rollout.
Go deeper
Want the full technical spec?
Download the deck or book a technical demo and we'll walk your team through the stack end-to-end.