Every AI
you’ve built
is blind.
Yours doesn’t
have to be

AI shouldn’t need another prompt to know something changed

Presence ambiently notices confusion, absence, attention changes, and shared visual context, then responds without interrupting work.

90 seconds · camera on · Google sign-in

She paused.
Looked away.
Sighed.
It noticed

Sit with everyone
who tries your product

Your best prospects get a founder watching them install, stepping in at the right moment, and learning what made the product difficult. Presence makes that available to everyone.

See how product trials work

After the Session

01Where they got stuck
02What it helped with
03What it left behind

It sees the person, not just the screen

Notices when someone goes quiet, looks away, or gets stuck. Then it speaks without being asked.

It keeps talking while it thinks

Slow tools and deep reasoning run in the background. No dead air.

Long conversations don't get more expensive

Minute 60 costs the same as minute 1. Meetings, debugging, onboarding, and support can keep going.

The prompt is not
the only channel

Presence notices what people don’t put into prompts.
It knows when to help, wait, or keep working.

How Presence works
You look confusedIt offers help
You step awayIt stops expecting a response
You returnIt knows you are back
You share somethingIt can react to what you are working on

It sees enough
to notice

Presence reads coarse signals such as present, looking away, screen shared, and engaged. It does not send a continuous video feed.

When the reasoning model needs to look, it receives a low-resolution frame. Higher resolution is available when the task calls for it. Frames are not retained in later conversation history.

The launch demo is not recorded. Other sessions require separate, session-specific consent before recording.

Read security and privacy

Choose one to see the setup.

Put Presence
in your product

Get your API key
WEB · THE WHOLE CLIENT INTEGRATION
<iframe src={session.embed_url} allow="camera; microphone; autoplay; display-capture" />

Read the installation guide

The parts are nameable.
Keeping them together is the work

Four engineering decisions make the difference between a stack of models and a live conversation that holds together.See every measurement

01 / TURN STATE

The face follows the conversation

Turn state, not audio amplitude, decides where it looks and when it shows processing. Mouth articulation is derived from the actual output audio, so it cannot drift from speech.

0.65 ms median
0.83 ms p95
Expression inference on CPU. 25 ms causal analysis window, 10 ms recurrent updates.
02 / PLAYBACK CLOCK

Audio is authoritative

Expression frames travel immediately before the PCM packet they describe, addressed in 16 kHz samples. If rendering stalls, stale poses are dropped instead of replayed late.

~155 msSource canvas to decoded WebRTC at 30 fps.
03 / ASYNC WORK

Tools do not stall the turn

The model dispatches work and keeps talking. Results arrive when the conversation is next idle; the video above shows the longer call returning while the session continues.

22.1 s and 9.7 sTwo concurrent think calls in the recorded demo session. Neither blocked a conversational turn.
04 / SPECULATION

Silence is spent generating

Presence starts generating at a provisional endpoint while it keeps listening. Resumed speech discards the work. Confirmed silence releases audio already made.

1.8–2.1 s p50
~1.2 s speculative
Warm full turn, microphone speech to first audio. A native-live baseline was faster on the same fixture: 1.054 s versus 1.955 s.
OBSERVATION STREAMRECORDED · 01:36
{
  "event": "pause",
  "signals": {
    "gaze": "away",
    "silence_ms": 2840
  },
  "note": "possible hesitation detected"
}
PRICINGPER OCCUPIED MINUTE
FreeFive 5-minute sessions. Google sign-in. No card.
$0.15$9 for 60 minutes.
$0.10$100 for 1,000 minutes.

Occupied time only. Credits never expire.