Voice Agents — Major Upgrade

We've just rolled out a major upgrade across the entire voice agent stack. If you've used Fedna AI voice before, you'll notice conversations flow noticeably better. If you haven't tried voice yet — now is the moment.

Smarter interruption handling

The agent now adapts to how you talk. It knows the difference between "actually—wait" (a real interruption) and "mm-hmm, yeah" (you agreeing while it talks). It holds its ground when you're just acknowledging, and yields instantly when you genuinely take over.

Adaptive turn detection

The agent decides when you're done speaking based on the pace and rhythm of the conversation — not a fixed timeout. Fast back-and-forth? Shorter pauses. Thoughtful answers? It waits. The net effect: no more "being talked over" and no more awkward silences.

Language lock hardened

On multilingual calls, language stays stable. The agent locks to the language you're speaking and doesn't accidentally drift into another one mid-sentence — even if you slip in a word or two from another language.

Call resilience

Calls now terminate cleanly when things go wrong upstream — silent lines, stuck sessions, provider hiccups — instead of sitting in a zombie state. Your team gets clean call logs with accurate outcomes.

First-turn context

The agent's opening line now knows whether it's greeting a brand-new caller or picking up a known customer. Returning customers get a recognition-aware opening; new callers get a proper greeting.

Why this matters

Voice is the hardest modality to get right. Every one of these improvements is a conversation artifact real callers notice — they just won't be able to articulate what changed. They'll just say the agent "sounds better" or "feels more natural."