Spring Update: Meet Naoma Agent V2, Powered by Conva

This spring we shipped the biggest upgrade to Naoma since launch: Agent V2, running on our own voice engine, Conva. It's now live for every active customer — and it makes our live AI demos faster, more natural to talk to, and far more controllable.

Here's what changed and why it matters.

Quick Takeaways

• Naoma Agent V2 is live for all customers — no migration needed on your side.
• It runs on Conva, our in-house, self-hosted voice engine.
• Lower latency — the agent responds closer to real time, so a demo feels like a conversation, not a walkie-talkie.
• Better interruption handling — buyers can cut in, change direction, and ask follow-ups mid-sentence, and the agent keeps up.
• Real control over the voice loop — we now own every stage, so behavior, guardrails, personalization, and languages are ours to tune.
• Built to scale — the new stack is more reliable under load and economically sustainable as demo volume grows.
• Proven in production — Naoma has now conducted 3,000+ live AI demos, with an average CSAT of 4.7/6.

Why we rebuilt the engine

A live demo is the most unforgiving thing an AI agent can do. A chatbot can take a second to think and no one notices. A voice agent that hesitates, talks over the buyer, or can't be interrupted breaks the illusion immediately — and a broken demo costs you the deal.

For our first generation, we assembled the voice pipeline on top of a third-party conversational stack. It got us live fast, but it was a black box: we couldn't fully control timing, turn-taking, or how the agent behaved in the edge cases that actually matter in a sales conversation. As we scaled past 3,000+ live demos, the limits became clear — both in experience and in cost.

So we rebuilt the core ourselves. Conva is a self-hosted engine that runs the entire voice loop — listening, understanding, and responding — under one roof we control end to end.

Lower latency

The single biggest driver of whether a voice demo feels human is response latency — the gap between a buyer finishing their sentence and the agent starting to answer. Owning the whole pipeline let us cut the hand-offs and overhead between stages, so V2 replies noticeably faster.

That speed compounds. The first 60 seconds of a demo decide whether a visitor stays or bounces, and snappy back-and-forth is what makes those seconds feel like a real conversation instead of a slideshow with a delay.

Better interruption handling

Real buyers don't wait politely for the agent to finish. They jump in: "skip that," "wait, go back," "does it integrate with our CRM?" V1 could stumble on that. V2 is built for it.

With direct control over voice activity detection and turn-taking, Agent V2 handles barge-in gracefully — it stops, listens, and adapts to where the buyer wants to go. The result is a demo that the prospect drives, which is exactly how the best human demos work.

Real control over the voice loop

This is the part that's invisible to buyers but matters most to us — and to you. Because Conva owns every stage of the pipeline, we can tune things a black box never let us touch:

• Behavior and guardrails — how the agent paces a demo, when it qualifies, when it hands off, and how it stays on-message.
• Personalization — adapting the demo to who's watching and where they came from.
• Languages — consistent quality across the 33 languages Naoma speaks.
• Reliability — fewer moving parts we don't control means fewer ways for a live demo to fail.

In short: the demo experience is now ours to improve, fast, instead of waiting on someone else's roadmap.

Built to scale

Self-hosting the engine wasn't only about quality. Running on a third-party conversational black box doesn't hold up — technically or economically — once you're delivering demos at volume across many customers. Conva gives us a stack that's stable under load and whose unit economics make sense as we grow, which is what lets us keep rolling V2 out to every customer without trade-offs.

And that rollout is done: Agent V2 is already running across every active Naoma deployment — part of the 3,000+ live AI demos Naoma has now conducted, at an average CSAT of 4.7/6. If you have Naoma on your site, you're on V2 today — nothing to install, nothing to migrate.

The bottom line

Agent V2 makes the thing buyers actually experience — the live demo — faster, more natural, and more reliable, while giving us full control to keep improving it. It's live for everyone now.

The best way to feel the difference is to talk to it. See a live AI demo, or explore Naoma pricing if you're ready to put V2 on your own site.