f[x]fluxions
newvui for iPhone is in beta

Teaching machines
to speak human.

We build the whole voice stack — recognition, reasoning and speech — from the weights up. Small enough to run on a phone. Cheap enough to build a business on.

Get the betaTalk to it
vui-nano, on an iPhone
35 seconds, sound on — every word is generated on the phone.

300Mparams

in vui-nano, speaking in real time on the iPhone GPU

100%on-device

recognition, the language model and the voice — nothing you say leaves the phone

Freeduring beta

iPhone with iOS 26+ · ~700 MB one-time download

§01 — vui for iPhone

A voice assistant that never leaves your phone.

vui runs speech recognition, the language model and a 300M-parameter voice entirely on the iPhone GPU. No server does the thinking, so nothing you say is sent anywhere.

  • Real voices

    Natural, expressive speech — hesitations, breaths and all — from a 300M model running on the iPhone GPU.

  • Talk to it

    Hold the mic and get a spoken reply. Recognition, the language model and the voice all run on the device.

  • Private by construction

    Your words never leave the phone. There is no server doing the thinking, because there is no server.

  • Works offline

    On a plane, on the tube, off-grid. Once the models are downloaded, vui needs no connection.

Get the betaTry it in the browser

§02 — API & pricing

The whole stack. One endpoint. One bill.

We train our own recognition, reasoning and voice, so nothing is stacked on top — no separate LLM bill, no TTS bill, no keys to bring. Billed by the second.

Voice agents, all-in
$0.10/ min
Transcription
$0.20/ hour
Text-to-speech
$9/ 1M chars (≈ $0.48 / audio hour)

Phone numbers for agents are bring-your-own or provided at cost.

See pricingRead the docs
render speech
curl -X POST https://api.fluxions.ai/vui/v1/tts \
  -H "Authorization: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "voice": "maeve.h8ff7e07da",
    "input": "[sigh] fine, I will say it one more time."
  }' \
  --output speech.wav
POST /vui/v1/tts. Square brackets are performed, not read: [sigh], [laugh], [gasp], [hesitate].

§03 — Phone agents

Answers the phone like your best receptionist.

Agents built on our API take bookings, triage call-outs and fill appointment books for real businesses — you can hear three of them here.

See pricing
a real call
preview · muted
  1. Call
  2. Call
  3. Call
Agents built on the Fluxions API, answering real business lines. Play one — or press Call and ring it yourself, no account needed.
what answers the phone

  1. akro — listens

    Speech to text — who is speaking, and the breaths, laughs and hesitations between words.

  2. LLM — thinks

    Decides what to say, and calls your tools — bookings, lookups, web search.

  3. vui — speaks

    Text to an expressive voice, streamed back while it is still being written.

All three stages are ours, so there is one bill.

§04 — Licensing

Run it on your own metal.

The same models, air-gapped in your infrastructure or on your own devices. Annual licence with SLAs and dedicated support.

Talk to usor say hello on Discord