◆ Solutions · Customer support
A face for your support agent.
Add a video surface to an existing support agent while keeping conversation logic, customer data, and escalation workflows in your stack.
◇ Problem
Why current solutions fall short.
- Text and voice interfaces do not provide the visual cues of a video conversation
- Pre-rendered avatar video cannot adapt to a live support exchange
- Seat-based licensing can be a poor fit for usage-based support volume
- Adding another interface should not require replacing the existing support stack
Stack-level cost: every escalated ticket is a real human-minute. The escalation rate is the lever.
◆ How Atlas fits
A passthrough video tier above your existing chat.
- Mount Atlas inside your existing chat widget as a 'Talk to support' button
- Audio into Atlas, video out; your stack owns STT, the support LLM, and TTS
- Customer can toggle video on / off without dropping the call
- Use visible speaking and idle states to make turn-taking clearer
- Keep escalation and human handoff inside your existing support workflow
◆ Deployment rationale
Realtime rendering can add a visual layer to an existing voice-first support agent.
- WebRTC works in any browser. No app install, no plugin.
- Bring your existing support LLM, prompts, and tools
- Per-second billing, so an unanswered button click costs zero
- Long-session support without re-auth or hand-off mid-call
◆ Integration shape
What it actually looks like in your codebase.
01
Mount the SDK
Drop @northmodellabs/atlas-react into your existing chat surface.
02
Wire your STT/LLM/TTS
Send transcribed user audio into your agent; pipe TTS output into an Atlas session.
03
Pick the avatar
Curated company-approved avatars or your own brand persona, single reference image per identity.
04
Ship
WebRTC stream renders in the user's browser. No per-seat license, no per-call cap.
◆ What to measure
The metrics that prove this is working.
Self-serve resolution rate
Compare video tier vs text-only chat for emotional or billing tickets, where chat usually plateaus.
Average handle time on escalation
Measure whether warm handoff with conversation context changes time to resolution.
Render cost / minute
$7/hr prorated to the second. Idle audio still bills session time; budget around real conversation length.
We don't publish customer numbers. The right comparison is against your existing baseline, your STT, your LLM, your voice, your traffic.
◇ FAQ for this use case
The questions that actually come up.
Does Atlas read our customer data?
No. Passthrough mode only sees audio frames and returns video frames. Your prompts, tools, and customer records never leave your stack. See /enterprise.
Can we use our own brand avatar?
Yes. One reference image per identity, and you can register multiple identities under the same account. See /safety for our acceptable use and public-figure policy.
What happens if our STT or LLM is slow?
Atlas renders frames as audio arrives, so the avatar stays in sync with whatever your stack outputs. Idle audio produces an idle avatar; session time remains billable while the session is active.
Ready to ship?
Email if you want async, book if you want a structured call. We'll follow up to discuss fit and requirements.