Product
Four things, done properly: build the agent, run the campaign, watch it live, understand the result.
Build
An agent is a prompt, a voice, and a set of tools.
You write what it should achieve and how it should sound. You pick the voice and the language, and decide whether callers can interrupt it. If it needs to do something real mid-call — look up an account, check availability, write a note back — you bind a tool to it.
- Prompt, voice and language in one screen
- Standard or premium performance profile
- Tools bound per agent, versioned so an edit can't silently change a live agent
Run
A campaign points an agent at a list and starts dialing.
Direction, timezone, calling window, days of the week, retry limits, caller IDs and the dispositions you want recorded — all on one screen with defaults already set. Consent is captured explicitly, and the server stamps the originating IP rather than trusting the browser.
- Outbound and inbound campaigns
- Dialing window and retry policy per campaign
- Explicit consent attestation on outbound
Watch
See what's happening while it's happening.
The live board shows every call in flight — the contact, the campaign, the agent, and how long it has been running — and rolls each one into the completed list the moment it ends. It refreshes on a short interval; it is a view of call state, not a stream of live audio.
- In-flight calls with elapsed time
- Completions in the last hour
- One click from a finished call into its recording
Understand
Then find out why it went that way.
Call review puts the recording next to the transcript, with the disposition that was recorded and the metadata around it.
- Recording and transcript in one workspace
- Search across transcript text
- Disposition history and call review
Inside 0.62 s
Where the wait actually goes.
Three things have to happen between you finishing and the agent starting. Neuvox times each one on every turn and shows you the split, because a single number tells you something is slow but never what.
- Hearing the end of your sentencespeech-to-text settles
- 180 ms
- Deciding what to sayfirst token from the model
- 260 ms
- Starting to say itfirst byte of audio
- 180 ms
- Total before you hear a word
- 620 ms
Illustrative figures from a playground session. Your own appear after every turn you speak.
Why it is hard
You cannot buy this back later.
Every part of the wait compounds: a slower model means the voice starts later, and a caller who waits twice starts talking over the agent. Most of the work in Neuvox goes into moving these three numbers down and keeping them there, which is why the product measures them on every single turn instead of once in a demo.
You can watch your own numbers move in the browser playground before a single call goes out.
Your telephony, your data.
Calls are placed through your own Twilio account and your own numbers. Credentials are stored encrypted server-side and are never sent back to the browser. Recordings stream through the application server rather than exposing storage URLs to the client.
Book a Demo