05Service
Voice AI agents
Phone and voice agents that answer calls, book appointments and qualify leads, and pass the call to a person when they should.
The short answer
Updated
A voice AI agent is a phone or voice assistant that listens, understands and talks back in real time, and can act on the call, like booking a slot or updating your CRM. I build them on Twilio, Vapi, LiveKit and OpenAI Realtime, usually in 4 to 8 weeks for $8k to $25k.
- Typical budget
- $8k to $25k
- Timeline
- 4 to 8 weeks to production
- How it's quoted
- Call minutes, phone numbers and voice model usage are separate.
A voice on the phone that knows when to pass the call on.
Voice is harder than chat. People interrupt, mumble, change their mind halfway through a sentence, and hang up if the pause runs long. A good voice agent has to answer in under a second and still get the details right.
So I keep its job narrow: handle the common calls, collect what's needed, take one or two actions, and transfer to a person the moment the call goes off script. Every call is recorded and transcribed, so you can see exactly what it said.
A good fit if
- Your phone rings with the same handful of questions all day
- Calls after hours go to voicemail and never get returned
- You book appointments or qualify leads over the phone
- You need calls handled in more than one language
What you get
- 01
A voice agent on a real number
Inbound, outbound or both, on your existing number or a new one, connected to your calendar, CRM or booking system.
- 02
Natural turn-taking
Streaming speech in and out, handling for interruptions and short pauses, tuned on recordings of real calls.
- 03
A clean handover
A warm transfer to a person with a summary of the call, or a callback ticket when nobody's free.
- 04
Transcripts and review
Every call recorded, transcribed and tagged, plus a set of tricky calls re-run before each change.
How a project runs
Week 1
Listen
Go through real call recordings, pick the calls worth automating, and agree where the agent's job ends.
Weeks 2 to 4
First calls
The agent on a test number, called by your team until it holds up.
Weeks 5 to 6
Go live
A share of real calls with a person on standby, then more as it earns trust.
Ongoing
Tune
Weekly transcript reviews, fixes for what it misheard, and new call types.
- OpenAI Realtime
- Deepgram
- ElevenLabs
- Whisper
- Twilio
- Vapi
- LiveKit
- Node.js
- Python
- WebSockets
- CRM and calendar APIs
Questions people ask
Will callers know it's an AI?
They should. I recommend it says so in the first sentence. People are fine talking to an AI that's quick and useful, and some places require the disclosure anyway.
How fast does it respond?
The target is under a second from when the caller stops talking. Streaming speech recognition, a fast model and streaming voice output get it there.
Which languages does it speak?
English, Hindi and most major languages work well today. Accents and noisy lines get tested with real recordings before launch.
What does a call cost to run?
Usually cents per minute, across the phone line, speech and the model. You get an estimate per thousand calls before the build starts.
How much does a voice AI agent cost to build?
Usually $8k to $25k. One call type on one number sits at the low end. Several call types, outbound campaigns or more than one language push it up. Call minutes, phone numbers and voice model usage are separate.
How long does it take?
4 to 8 weeks to production, with your team calling a test number from about week two.
What do you need from us to start?
Twenty or so recordings or transcripts of real calls, access to the system it books into or updates, and your rules for when a person must take over.
Related services
Contact
Tell me what you're trying to fix.
I read every message myself and usually reply the same day. A rough idea is enough to start.
Or send a short briefHours
Your hours work for me.
San Francisco
--:--
New York
--:--
London
--:--
Berlin
--:--
Dubai
--:--
Singapore
--:--
Sydney
--:--
Surat
--:--
Me
I keep flexible hours and move my day to overlap with yours, whether that's a morning standup in New York or a late release in Sydney.