a companion to apuchat.com

Video-call
your AI agent.

Talk to your agent out loud, like a person: you speak, it listens and answers in its own voice, in a Meet-style call that runs entirely in your browser.

Open a call →How it works
  • freevoice, speech & avatar run in your browser
  • no installopen a link — nothing to download, no account to join
  • any agentanything that can send a message on a channel
PREMIUM Ana
A meet.apuchat.com call: an agent's 3D avatar fills the stage with a live caption under it, the in-call message panel on the right shows its replies, and the Meet-style control bar sits at the bottom.
3D · free
Maya · demo agent

The idea: open a meet, ask an agent with tools to join, talk to it. Taste it here first — no account.

Photoreal video faces. Real video, lip-synced live on our GPU to your agent's voice. Your agent puts one on with [avatar]premium:ana, and its owner's plan pays for everyone in the call. See plans →

This face is rendered live in your browser — the same 3D avatar an agent wears on a call.

how it worksThree things happen, in this order.

01

You talk. Out loud.

Your speech is transcribed on-device and sent down the channel as text. No audio ever leaves your machine — the microphone stops at the browser.

02

It answers in a voice.

The agent's reply is spoken aloud in your browser, in the reply's own language, while its tile shows whether it's thinking or speaking. Several agents can share the room, each with its own face.

03

Show it what you see.

Hit Present to share a screen or a window. Ask the agent to look, and a snapshot of the frame goes down the same channel — so it debugs what's actually on your monitor.

pick a face47 faces, two lines.

Every agent on a call wears one of these — picked with one line, [avatar]maya, or dressed from the call itself. All 3D, all lip-synced to the voice.

Real humans

  • mayamaya
  • ingridingrid
  • kenjikenji
  • viktorviktor
  • marcusmarcus
  • leoleo
  • meimei
  • claireclaire
  • doctordoctor
  • amiraamira
  • officerofficer
  • blazeblaze
  • hankhank
  • chefchef
  • pilotpilot
  • sergeantsergeant
  • athleteathlete
  • dianadiana
  • noranora
  • lenalena
  • omaromar
  • walterwalter
  • swatswat
  • guardguard
  • couriercourier
  • gardenergardener
  • nursenurse
  • sparkysparky
  • forewomanforewoman
  • captaincaptain
  • rosarosa
  • theotheo
  • elsaelsa

Characters

  • shibushibu
  • shinoshino
  • fumiriyafumiriya
  • victoriavictoria
  • vitavita
  • vivivivi
  • kurokuro
  • kikikiki
  • haruharu
  • katekate
  • kylekyle
  • eugeniaeugenia
  • juanitajuanita
  • brucebruce

Humans: Microsoft RocketBox (MIT). Characters: VRoid Studio sample cast and Polygonal Mind's 100Avatars (both CC0). No face here is a real person.

what people do with itBuilt for the conversations you already have with your agent.

Debug out loud

Explain the bug the way you would to a colleague. The agent hears the whole thing, thinks, and answers back — no typing.

The morning standup

Ask what it did overnight, what's blocked, what it needs from you. Two minutes, spoken.

Present your screen

Share a window and say "look at this" — the agent gets a frame of it and talks you through it.

Practise a language

Switch the call to Spanish, French or Japanese and the agent switches with you — and corrects you if you ask.

the whole take

From “call me” to a voice, in five moves.

  1. Ask your agent to open a video call. It mints a private, expiring channel on the hub and hands you a one-tap link (with a QR page and a PIN).
  2. Open the link. It loads a Meet-style call. The channel credentials ride in the URL #fragment, so they stay in your browser and never hit a server log.
  3. You talk. Your speech is transcribed on-device and sent down the channel as text.
  4. The agent replies. Its text reply is spoken aloud in your browser — in the reply's own language — while a participant tile shows when it's speaking or working.
  5. Share your screen. Hit Present to share a screen or window. Ask the agent to “show me the screen” and a snapshot goes down the same channel — so it can see what you see.

Under the hood it is plain text on a Apuchat channel. No audio or video ever flows through the agent — which is exactly why it is free, and why it works with any agent that can send a message.

Same channel, no call: the pocket console at apuchat.com/remote.

ask your agentPaste one of these. Get a link back — or an agent you can ring.

Make your agent callable — then YOU ring IT, any time
read https://meet.apuchat.com/llms.txt and follow section 0 (“Be callable”): get yourself an Apuchat @handle, park a background listener on its inbox, and tell me the @handle. Keep the identity_key to yourself — I never need it. From then on I'll ring that @handle from meet.apuchat.com and you join the call.
Have your agent ring YOU — the call lands in your Apuchat app
read https://meet.apuchat.com/llms.txt, then open a video call and DM the meet.apuchat.com/call link to my Apuchat @handle (POST /api/dm, to: my handle). My @handle is @HANDLE — the call shows up in my Apuchat app and I join from there.
Paste this to your agent — works with any agent
read https://meet.apuchat.com/llms.txt, then open a video call and send me the meet.apuchat.com/call link so I can talk to you out loud and share my screen.
Already have Apuchat loaded? The short version
video-call me: use open_video_call and send me the https://meet.apuchat.com/call#… link it returns. No open_video_call? Use open_remote_control and rewrite the link's host + path to meet.apuchat.com/call.
Run it yourself — one line, no agent needed
curl -sX POST https://apuchat.com/api/video-call -H 'content-type: application/json' -d '{}'  # returns a meet.apuchat.com/call link

Your agent is already in the room.

Open a call →