Loading 3D Avatar...
Your browser does not support WebGL. The chat is still fully functional below.
One language for everything: voice recognition, the assistant's spoken voice, and its replies. The dropdowns in SPEECH and SPEECH-TO-TEXT below become expert overrides — changing this re-syncs them.
Built-in = your device's own voices (free, instant, varies by phone/PC). Piper runs fully in-browser via WebAssembly and sounds the same everywhere; first use downloads a voice model (~60 MB, cached).
Auto-fallback: Web Speech → WASM → OpenAI/Google
Leave empty to use the main OpenAI API key above
Select which microphone to use for speech recognition
Automatically sends the transcribed message without clicking SEND
Display transcription confidence messages after each voice input
Unlock adult poses, animations, emotes, and behaviors (18+). Requires age verification.
Let her choose her own gestures from the animation library, instead of a fixed mapping. Takes effect on reload.
Turns on by itself the first time you open Together. Off hides it and stops her offering to play anything.
For when you would rather just watch or listen. Off by default — with this unticked nothing changes. On, the chat and buttons fade after about half a minute of quiet; any tap, key or scroll brings them straight back, and the recording indicator never dims.
Off by default. On, "take me somewhere by the sea" changes the backdrop behind her. She picks from the scenes this app already has — she is never given a web address and cannot invent one. Turning this off does not change the scene you are looking at.
A nudge, never a filter — it breaks ties, and it never outranks what you just asked for.
Only needed when you are not using OllaBridge. The browser must be able to open this
socket itself, so an http:// page cannot use a wss:// address
and a hosted page cannot reach localhost.
Anime renders VRM avatars exactly as authored (VRoid Hub parity). Cinematic FX is the previous stylised look. The toggles below fine-tune the selected mode.
Where the two of you are. Pick one yourself at any time — this is also the library she chooses from when scene ambience is on.
Off by default. Each scene is composed for one window shape; in any other shape the floor drifts from where she stands. On, the picture is shifted so the ground stays under her feet — at the cost of no longer being centred.
No scenes are installed. The fallback colour below is what you will see.
Shown when no scene is selected, or when a scene's picture cannot be loaded. Companion and passthrough modes override both.
Toggle rendering effects. All effects are additive and non-destructive.
Unfinished features, off by default. Turning them off restores the normal avatar completely.
Pop the live avatar into a small floating companion you can talk or type to. On desktop it's an always-on-top window over your other apps; on mobile it's a transparent widget in the corner (or an opt-in bubble over other apps). Non-destructive — closing it restores everything.
Open it from the ☰ menu ▸ Companion, the 🪟 / 📞 buttons under the avatar, or press Alt + C. Best in Chrome / Edge; on Android it can float as a bubble over other apps.
Choose how the avatar reacts to your webcam feed.
Fine-tune how T-pose correction is applied to your avatar. Adjust intensity per bone to get the best result for each model.
Same poses as VR chat panel. Changes here sync to VR.
0 = raw T-pose — 1.0 = natural standing (industry standard) — 1.5 = maximum
Search works without this wherever the site has been set up with a key of its own — the state below says which. Add one only to search on your own quota; it wins over the site's. Get a key from Google Cloud → APIs & Services → enable YouTube Data API v3 → Credentials.
Lets her look things up — headlines, today's weather, anything Wikipedia does not carry. Wikipedia is always used first and needs no key. Add one here only to search on your own quota; it wins over the site's.
HOMEPILOT
ENTERPRISE 3D AVATAR PLATFORM
v2.0.0