The 3D characters Polybot and DinoKid under a stage light, next to the word bibabibo

Live characters, rendered in your browser

Voice assistant · 3D character

Talk out loud with a 3D character

It hears you and answers aloud, right away. The character lip-syncs, blinks and remembers you between conversations.

Just talk, like you would to a person

No typing, no push-to-talk. The character hears you out, thinks for a beat and answers aloud. You can cut in any time.

Lips that follow every sound

Mouth shapes track the voice, eyes blink and wander naturally, a hand goes to the ear while listening and to the chin while thinking.

Pick a character, or bring your own

Built on the open VRM standard: choose DinoKid or Polybot, or point the admin page at any VRM file.

Remembers you, conversation after conversation

Your name, what you like, what you left unfinished: each character keeps its own memory and brings it up next time.

Sits in on your meetings

Tell it to “just listen”: it stays quiet, keeps a record of the meeting, and only speaks up when someone calls it by name.

Pay for what you use

Model cost is metered per turn, plus a 10% management and development fee. No plans, no subscription.

Features

What bibabibo does today, not a roadmap.

  • Real-time conversation

    One live connection to a realtime voice model: it hears you and answers aloud straight away, and you can interrupt mid-sentence.

  • A character that feels alive

    Lip-sync to the voice, blinks, glances, breathing; a hand to the ear while listening, to the chin while thinking.

  • Per-character memory

    Each character remembers what you told it across sessions and brings it up next time.

  • Listen-only mode

    In a meeting it quietly keeps every line and only replies when called by name.

  • Live transcripts

    The admin page shows every line from speaker and character while the session is still running.

  • Cost down to each turn

    Every model call is logged and priced in USD, broken down by user, agent and session.

Runs in the web browser and on Android.

What it is for

One character, many roles.

  • English speaking practice

    Daily conversation with a patient partner that remembers how far you have got.

  • Personal assistant

    Quick questions, suggestions, thinking out loud. Hands-free, no typing.

  • Meeting notes

    Put the phone on the table, switch on listen-only mode, call its name when you need a recap.

  • A friendly voice at home

    A likeable character the whole family can chat with every day.

Characters

Two characters ship with it, both CC0-licensed VRM models.

Polybot, a robot with a smiling screen face

Polybot

A robot with a smiling screen for a face. Light (1 MB) and quick to load, even on phones.

100Avatars #051 · Polygonal Mind · CC0

DinoKid wearing a dinosaur onesie

DinoKid

A character in a dinosaur onesie, friendly to every age.

100Avatars #045 · Polygonal Mind · CC0

Have your own model? The admin page can point at any VRM file, and framing adapts to the character’s height.

How it works

From creating a character to reading the bill.

  1. Create an agent

    On the admin page: name, personality, voice and character.

  2. Open the app and talk

    Sign in on the web or Android, tap the round button and just speak.

  3. Heard and answered instantly

    The realtime model listens to your voice directly and replies aloud; the character lip-syncs along.

  4. Review and remember

    The conversation appears live on the admin page; after the session, the character updates its memory.

  5. See every cost

    Each turn is logged with its tokens and price, plus the 10% platform fee.

Pricing

Pay as you go.

  • You pay exactly the model cost of what you use, metered per turn from the token counts the provider reports, times the list price.
  • Plus a 10% platform management and development fee. No plans, no monthly subscription.
  • Every turn shows up on your Cost page, down to each individual model call.
gpt-realtime-2.1 list prices
KindUSD per 1M tokens
Audio in32
Audio out64
Text in4
Text out24
Cached input0.4
Source: OpenAI list prices, read 2026-09-25. Prices may change with the provider.
Example: a 2-turn session with memory, measured 2026-09-25
ItemUSD
Turn 1 (realtime)$0.0139
Transcription (realtime)$0.0012
Turn 2 (realtime)$0.0183
Memory (LLM + embedding)$0.0203
Model cost$0.0537
Platform fee 10%$0.0054
Total$0.0590

FAQ

What is bibabibo?

An AI assistant you talk to by voice, with a 3D character on screen. You speak, the character listens, answers aloud straight away, and remembers you between conversations.

Which devices does it run on?

In the web browser (desktop or phone) and as an Android app.

Does it speak Vietnamese?

Yes. Characters can talk in Vietnamese and English; each character’s personality and voice are set on the admin page.

How is it billed?

Pay as you go: the actual model cost of each turn, plus a 10% management and development fee. No plans, no subscription. For example, a 2-turn session with memory came to $0.0590.

Where is conversation data stored?

Conversations are stored on bibabibo’s server so the account owner can review them on the admin page. Memory is kept separately for each character.

How do I get an account?

Accounts are currently issued by an administrator. If you already have one, open the app and sign in.

Ready to talk?

Already have an account? Open the app and start talking.