How it works
From your voice to a spoken answer
A single low-latency loop, four best-in-class layers. Here's exactly what happens between you speaking and the panel speaking back.
A single low-latency loop, four best-in-class layers. Here's exactly what happens between you speaking and the panel speaking back.
The browser opens a secure, real-time connection and streams your microphone. No phone number, no dial-in — it all happens in the tab.
Your speech is streamed to text the moment you talk, so the panel can start reasoning before you've finished your sentence.
Each persona declares its expertise. A router reads the turn and hands it to the best-fit specialist — or the generalist host when nothing else clearly fits.
The chosen expert thinks with the AI model you've assigned it, pulling from your uploaded documents and the live web when needed — grounded, not guessed. Different experts can use different models for many perspectives.
The reply is voiced in real time with natural, low-latency speech in that expert's configured voice and pace.
The conversation is transcribed and saved, and expert cards stay live on screen so you can tune any voice mid-call.
Behind the scenes, voicer is one hosted service doing two jobs — the web app and the live voice agent — which keeps it fast and reliable. You just sign in and talk.
The web app serves the dashboard, admin and chat pages, the documents, search, notes, contacts and folders API, sign-in, and the voice session endpoints.
A live voice agent holds a persistent connection and is dispatched automatically into every new room to join the conversation and talk.
you speak ──▶ transcribe ──▶ router picks expert ──▶ your model reasons ──▶ voice reply
(grounded by your documents + the live web, all in the browser)
A deeper write-up sits behind a password: every feature in more detail, the architecture, the providers, how the models are chosen and metered, and how the whole thing is tested. Ask Paul for the password.
The fastest way to understand voicer is to talk to it.