The live 3D avatar
The avatar is the face of your agent, rendered in real time. Its eyes, expression and posture show exactly what your agent is doing — and it can read replies out loud.
The states
The avatar's look changes with what your agent is doing right now:
- Idle — resting and ready, waiting for your next move.
- Thinking — making sense of your intent and planning the approach.
- Working — hands on, calling tools to get the job done.
- Blocked — hit a snag and needs your confirmation or a hand.
The avatar speaks
The avatar reads replies out loud using your browser's built-in speech. Turn on Auto-read answers in the voice settings, or tap Read aloud on any message to hear it on demand.
- Its mouth moves with the voice, so you can see it talking as it reads.
- Reading is built in — no extra account or subscription needed.
Choose or install a voice
The voice list comes from your operating system. A web page can't install voices itself — it only lists the ones already installed — so to add a voice you install it at the OS level, then come back to the voice panel.
macOS
- Open System Settings → Accessibility → Spoken Content → System voice.
- Tap Manage voices and download the voice you want — Yue is a high-quality built-in Chinese voice.
- Reopen the voice panel (or refresh the page) so the new voice appears in the list.
Windows
- Open Settings → Time & language → Speech → Manage voices.
- Tap Add voices and download the language or voice you want.
- Reopen the voice panel (or refresh the page) to see the new voice.
iOS / iPadOS
- Browsers on iOS only expose the built-in voices — a downloaded enhanced voice like Yue won't appear in the list, even after installing it. Install it on a Mac or Windows PC to use it there.
Changed voices? If a voice you'd saved is no longer available, the panel falls back to Auto (system) — just pick a voice again.
Make your avatar your own
The avatar is fully customizable. Drop in your own VRM model (or keep the default) and the panel parses every material on it, grouped by part — hair, body, eyes, face, clothes, backpack. Any texture can be recolored, re-patterned, or replaced with your own image. The loop is repeatable: pull the original texture off the model → have the agent generate a new asset → swap it back on. It all happens in your browser, with no file export.
Step 1 — extract the original asset
Open the avatar panel (Customize → Textures). Every texture on the model is grouped by part: hair, body, eyes, face, clothes/gear, backpack.
- Click a texture's thumbnail to open it at full resolution — save or screenshot the original from there.
- The smoother path is Add to board: it puts the texture's full image onto the image board, which travels with your next message as a reference for the agent.
- Textures are the model's finished images — they go onto the board uncompressed with transparency intact, so the agent sees exactly what's on the model.
Step 2 — generate a new asset with the agent
With the original texture on the board, just say what you want — a new color, a new pattern, an added motif, anything.
- The agent's new image lands in the Create gallery and stays in the conversation.
- Not happy? Say so and iterate, same as any image edit — no need to nail it first try.
- No export needed — the next step picks straight from the gallery.
Step 3 — swap it onto the avatar
Back in the texture section, target the same texture and pick one of three ways to replace it:
- From conversation — open the gallery icon and apply the image you just generated.
- <b>Upload</b> — use a local file, or one you drew yourself.
- Swatches / custom color — one-click recolor, best when you only want to change the palette.
- Changes apply live; <b>Reset</b> restores a single texture and <b>Reset textures</b> restores all. A new image keeps the original texture's transparency outline, so you only supply the pattern itself — no manual cut-out.
Changes stay in your own browser (IndexedDB) — nothing is uploaded or synced across devices.