Agent Box
Premium ✦
Call it. Talk to it.
A complete voice AI assistant. It answers the call, understands, and replies out loud — you can interrupt it mid-sentence like a real conversation.
NEW Sign up now — get 100,000 characters free
Excellence in Inference Engineering.
Foundational AI that thinks — and runs — differently.
Text-to-speech & speech-to-text · REST + WebSocket · 12 natural voices · priced per character
LIVE
TTS & STT · Voice API
Two calls — text to speech, speech to text
# text → speech
curl api.quickdial.ai/v1/tts \
-d '{"text":"Hi","voice":"azelma"}' -o hi.wav
# speech → text
curl api.quickdial.ai/v1/stt -F audio=@hi.wav LIVE
Sophie · Agent Box
Answering for Acme Salon — 24/7
Hi! Thanks for calling Acme Salon — how can I help?
Do you have anything free tomorrow afternoon?
Yes — 2:30 or 4:00. Want me to book you in?
$0.0098/min
Agent Box, all-in
Under a cent a minute — 40× cheaper than Vapi
245 ms
Agent answers back
Voice-to-voice p95 — ~3× faster than Retell & Vapi
24/7
Never misses a call
Phone + website, every hour of every day
$0.0018/1K
Cheapest API
STT & TTS — a fraction of a cent per 1,000 characters
<75 ms
Fastest first audio
Real-time, streams as it's generated
99.99%
Always on
Four-nines uptime · unlimited concurrency
Three products, one platform
Move between text and voice, instantly.
Text to Speech
Type it. Hear it.
12 natural English voices, streamed back in real time as you generate — WAV or Opus.
Speech to Text
Speak it. Read it.
Upload or stream audio and get an accurate transcript with word-level timestamps — powered by whisper.cpp.
Live demo
Try it right here — free.
Free daily demo · up to 300 chars
Agent Box
A voice agent that answers the phone.
Listens, understands and answers out loud in real time. The moment you stop speaking it runs the full journey — speech-to-text, language model, speech synthesis — and the first words come back in about 240 ms. Interrupt it and it stops to listen, just like a person.
Get your own agent →Point it at your website, pick your tools, and it answers your calls — live at agentbox.quickdial.ai.
Option 1
Talk from your browser
Tap, allow the mic, then just speak.
You said
Agent
That was Agent Box.
From $8/month — live on your site in 60 seconds.
Sign up to keep going →
or call +1 (424) 567-8978
Chrome, Edge or Safari · mic audio is processed live and never stored.
Option 2
Call it or scan
Ring the demo line from any phone — the same box picks up.
US number · standard carrier rates apply
The price gap
Under a cent a minute. Everything included.
What you actually pay per connected minute, with speech-to-text, the language model and speech synthesis all included. Other platforms quote an orchestration fee and bill the models on top.
Cost per connected minute
all-in: speech-to-text + language model + speech
Vapi
$0.05 headline + GPT-4o + ElevenLabs
$0.400
40× more
Retell
$0.07 headline + frontier models
$0.310
31× more
Bolna
₹5.52/min fee + frontier models
$0.220
22× more
ElevenLabs Agents
frontier models · $0.16 during bursts
$0.120
12× more
Quickdial Agent Box
speech-to-text, language model and speech — one line item
$0.0098
40× cheaper
Competitor figures run their frontier models at published list prices — the platform fee plus their top speech-to-text, language model and voice (for example GPT-4o with ElevenLabs Multilingual). Smaller-model tiers land nearer $0.10–$0.15. Telephony is passed through at carrier cost on every platform, including ours.
The latency gap
Answers before the pause gets awkward.
Every platform below is measured the same way: from the moment the caller stops speaking to the first sound of the reply — the complete speech-to-text → language model → speech synthesis journey.
Voice-to-voice response
end of speech → first audio · p95
Vapi (optimized)
p95 · 720 ms median · 500 production calls
1,050 ms
2.9× slower
Retell
p95 · 680 ms median · 500 production calls
920 ms
2.5× slower
Quickdial Agent Box
p95 · measured on the live demo box
245 ms
~3× faster
Competitor figures: independent study of 500 production calls per platform, March 2026. They chain speech-to-text, a language model and speech synthesis across separate cloud vendors, so every turn pays several network round-trips; Agent Box runs all three together, with nothing between them. Our figures come from 16 consecutive turns on the live demo box you can try above — a smaller sample than their 500 calls, but the same metric. The spread is tight (231–419 ms) because there is no network or vendor queue inside a turn. ElevenLabs publishes 75 ms, but that is speech synthesis alone, not the full conversation loop, so it is not comparable here.
Why Quickdial
Built to be cheap, fast, and boring to operate.
⚡
Real-time streaming
REST and WebSocket both stream audio as it's generated — the first words play before the sentence finishes.
¢
Up to 32× cheaper
The cheapest voice API on the planet — up to 32× cheaper than the big providers, optimized end to end. Pay only for what you use, no tax on scale.
🎙
12 natural voices
Male & female, conversational & reading — expressive English voices for every use case.
📝
Speech-to-text too
The other direction, even cheaper. Upload or stream audio, get a timestamped transcript.
∞
Unlimited concurrency
Stateless workers autoscale horizontally — handle one request or a million, no capacity planning.
🛡
99.99% uptime
Four nines of availability, per-key rate limits, and metered billing you can trust to the fourth decimal.
The price gap
Up to 40× cheaper, every way.
Compared against each provider's frontier models, at published list prices.
Agent pricing is all-in per connected minute — speech-to-text, language model and speech synthesis together. Other platforms headline an orchestration fee only ($0.05–$0.07) and bill the models on top; the figures here are their delivered cost running frontier models.
Speech-to-Text
per 1,000 characters
Google / Azure (latest)
$0.016/1K
8.9× more
AssemblyAI (Universal)
$0.0062/1K
3.4× more
OpenAI (Whisper)
$0.0060/1K
3.3× more
Deepgram (Nova-3)
$0.0043/1K
2.4× more
Quickdial
$0.0018/1K
—
Voice agents · all-in
per connected minute
Vapi (frontier models)
$0.400/min
40× more
Retell (frontier models)
$0.310/min
31× more
Bolna (frontier models)
$0.220/min
22× more
ElevenLabs Agents (frontier)
$0.120/min
12× more
Quickdial Agent Box
$0.0098/min
40× cheaper
Text-to-Speech
per 1,000 characters
ElevenLabs (Multilingual v2)
$0.090/1K
32× more
Sarvam (Bulbul v3)
$0.031/1K
11× more
OpenAI (tts-1-hd)
$0.027/1K
10× more
Google / Azure (Neural2 · HD)
$0.016/1K
6× more
Quickdial (Opus)
$0.0028/1K
—
Each competitor shown at its flagship model on published list pricing. Speech tables are $ per 1,000 characters; agent pricing is $ per connected minute. For directional comparison only.
Trusted by builders
Nobody misses a call any more.
What founders and operators say after putting Agent Box on their phone line and website.
✦ Agent Box
Agent Box gave every one of our merchants a receptionist overnight. It answers on the first ring, books the job, and the whole thing costs less per month than a single missed booking used to.
AK
Ankur Kapoor
Founder & CEO · Easy Dial AI Inc
✦ Agent Box
We rolled Agent Box out across our clients and the maths stopped being an argument — under a cent a minute, all-in. Same conversation quality, a fraction of what the big agent platforms wanted.
PC
Founder · Infinisync Consulting Private Limited
✦ Agent Box
Agent Box now answers every call that used to go to voicemail. It booked appointments on the first weekend — for less than we were paying for one afternoon of reception cover. Callers genuinely can't tell it isn't a person.
VS
Co-founder & CTO · PlatterHub AI
✦ Agent Box
Agent Box is the first voice agent we've seen answer in a quarter of a second at under a cent a minute. Collapsing that cost curve is exactly the kind of foundational infrastructure we look to back.
AS
Managing Partner · D26 Ventures
Pricing
Start free. Pay only for what you use.
Voice engines a fraction of a cent per 1,000 characters — with 100,000 characters free (~100 min of speech) on the voice API when you sign up, no monthly minimum. Agent Box, the premium agent, from $8/month.
✦ Premium
Agent Box
$0.0098
/min Under a cent per minute — on Premium
Starter
1.96¢ per minute
$8/mo
Premium
$0.0098 per minute
$18/mo
- ✓ Answers web & phone 24/7
- ✓ Books appointments, transfers to humans
- ✓ Learns your business from your website
- ✓ Live in 60 seconds — one script tag
Speech-to-Text
$0.0018
/1K chars whisper.cpp · word-level timestamps
- ✓ Accurate multilingual transcription
- ✓ Word-level timestamps
- ✓ Upload or stream audio live
- ✓ REST + WebSocket
Text-to-Speech
$0.0028
/1K chars Opus format · WAV (lossless) at $0.0045 / 1K
- ✓ 12 natural English voices
- ✓ Real-time streaming — WAV or Opus
- ✓ First audio in under 75 ms
- ✓ REST + WebSocket
What every minute buys
Not just a low price — the agent and the platform behind it, on every call.
245ms
Answers back
Voice-to-voice p95 — ~3× faster than Retell & Vapi
40×
Cheaper than Vapi
All-in per connected minute, no model bill on top
24/7
Phone + website
Books, answers and transfers — never misses a call
99.99%
Four-nines uptime
Unlimited concurrency, no capacity planning
Create your API key
100,000 characters free (~100 min) on the voice API · no card required · cancel anytime
Developers
Speech in one request.
One key, one endpoint per direction. Plain REST when you just need the bytes, WebSocket when you want to stream as it's generated — from any language.
- 🔑 One Bearer key — Same key for TTS, STT and both stream endpoints.
- ⚡ Streaming built in — /v1/tts/stream and /v1/stt/stream over WebSocket.
- 🧩 Any stack — curl, fetch or the SDK — no special client needed.
- 📈 Metered to 4 decimals — Per-second billing you can reconcile to the cent.
quickstart.sh
# Text → speech, streamed to a file
curl https://api.quickdial.ai/v1/tts \
-H "Authorization: Bearer qtts_live_your_key" \
-d '{"text":"Hello world","voice":"azelma"}' \
--output hello.wav
# Speech → text, with word timestamps
curl https://api.quickdial.ai/v1/stt \
-H "Authorization: Bearer qtts_live_your_key" \
-F audio=@speech.wavIntegrations
Drops into your voice-agent stack.
Official plugins for the frameworks you already build on — swap in Quickdial TTS & STT with one line, keep the rest of your pipeline.
LiveKit Agents
Python v0.1.4 · Node v0.1.4
Real-time STT & TTS as native LiveKit plugins for voice agents.
pip install livekit-plugins-quickdial
View plugin → Pipecat
Python v0.1.0
Quickdial STT & TTS services for Pipecat voice pipelines.
pip install pipecat-quickdial
View plugin → Questions, answered.
What is Agent Box? +
A complete AI voice agent for your business. It learns what you do from your own website, then answers your phone line and your web visitors 24/7 — booking appointments, answering questions and passing the tricky ones to a human. You can interrupt it mid-sentence, like a real conversation.
How much does Agent Box cost? +
Starter is $8/month at 1.96¢ per connected minute; Premium is $18/month at $0.0098 per minute — under a cent. That price is all-in: speech-to-text, the language model and speech synthesis together, with no orchestration fee and no per-token bill on top. Telephony is passed through at carrier cost.
How long does it take to set up? +
About a minute. Sign in with Google, give it your website, pick what it should learn and which tools it may use — then paste one script tag on your site and/or point a phone number at it. No backend to build.
How fast does the agent reply? +
245 ms at p95 from the moment you stop speaking to the first sound of the reply — roughly 3× faster than Retell or Vapi. Every stage runs co-located in one process, so there is no network hop inside a turn.
How much does the voice API cost? +
Just $0.0018 per 1,000 characters for speech-to-text and $0.0028 for text-to-speech (Opus; WAV is $0.0045 for the larger lossless payload). Pay-as-you-go, 100,000 free characters (~100 min) on sign-up, no monthly minimum.
Is it really real-time? +
Yes. Both REST and WebSocket stream audio as it's generated, so the first words play before the sentence finishes — with no latency penalty versus the raw engine.
How does it scale? +
Stateless workers autoscale horizontally for effectively unlimited concurrency, backed by a 99.99% uptime target and per-key rate limits.
Which voices? +
12 natural English voices — male and female, conversational and reading — for TTS; accurate multilingual transcription for STT.
The bigger picture
Voice is just the beginning.
Quickdial is building foundational AI — starting with the cheapest, fastest voice on the planet. More modalities and solutions are on the way.
✓ Text-to-Speech ✓ Speech-to-Text ✓ Agent Box More modalities · soon New solutions · soon
Ship voice today, for a fraction of a cent.
100,000 characters free, no card required. Cheapest, fastest, always on.