Skip to content

Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.

5.1 KB
Raw Blame

Make your first agent

This guide walks you through creating a 3D AI agent in the browser — no code, no 3D software, no crypto wallet required.

Time: about 5 minutes
What you'll have at the end: a live 3D AI character you can share or embed on any website


Step 1 — Open the wizard

Go to three.ws/start. This is the new-agent wizard.

You can walk through the whole setup before signing in: the avatar, the name, the brain, and the skills are kept in the tab. The wizard asks you to sign in (or create a free account) at the publish step, then picks up exactly where you left off. See Do I need crypto? if you're not sure which sign-in option to choose.


Step 2 — Pick an avatar (the body)

The first step of the wizard asks you to choose a body for your agent.

Your options:

  • Browse the gallery — hundreds of ready-made 3D characters, from realistic to stylized
  • Take a selfie — upload a photo and get a 3D version of yourself in about a minute at /create/selfie
  • Upload a GLB — any rigged glTF/GLB file works. Drag it into the editor at /app or import from a URL

Not sure what an avatar is vs. what an agent is? Read Agents vs. Avatars.


Step 3 — Name your agent and describe its personality

Give your agent a name and a short description of who it is. For example:

"You are Aria, a helpful product guide for Acme Corp. You speak in a friendly, concise tone. You know everything about our products and can answer questions about pricing, features, and how to get started."

This description is your agent's system prompt — the instructions it follows. You can always edit it later.


Step 4 — Enable skills (optional)

Skills are extra capabilities you can give your agent. The wizard offers:

Skill What it does
Memory The agent remembers what you tell it across sessions (on by default)
Deep thinking The agent reasons step by step before answering (on by default)
Web research The agent can look things up in real time

There is also an optional crypto group (pump.fun monitoring/trading, a Solana wallet, and x402 pay-per-call) you can reveal and enable if you want an on-chain agent; skip it entirely otherwise.

You can add or remove skills at any time from the agent's edit page. Voice (microphone input and spoken replies) isn't a wizard skill: you configure it later in the Voice panel of the agent editor.


Step 5 — Publish and get your embed

Once you hit Publish, your agent gets a live URL: something like three.ws/agents/abc123 (plus a friendly handle URL like three.ws/@aria once you reserve a handle).

The wizard also gives you an embed snippet: a few lines of HTML you can paste into any webpage to make the agent appear there. You don't need to know HTML to do this — just copy and paste.

For more on where and how to embed, continue to Share & embed →


Editing your agent later

Everything is editable after publishing. Go to your agent's page and click Edit. You can change:

  • The avatar (swap the body without losing the personality)
  • The system prompt / personality
  • Which skills are active
  • The agent's name, description, and thumbnail

Changes take effect immediately — no republishing needed for most settings.


How the avatar step hands back to the wizard

Step 1 of the wizard sends you to one of the avatar pages (/create/selfie, /create, /create/prompt, /create/studio) and expects the finished avatar back. Those pages share one small contract, implemented in src/shared/wizard-return.js:

  1. The wizard links out with ?wizard=1&next=/start?from=selfie.
  2. The first page in the chain calls captureWizardReturn(), which keeps the next path in sessionStorage and removes it from the address bar. Only a same-origin path under /start is ever stored; anything else is ignored.
  3. Whichever page finally saves the avatar calls pendingWizardReturn(). When a return is pending, it offers "Use this avatar and continue setup" (the selfie and prompt pages) or goes straight back (the review and studio pages) via returnToWizard({ avatarId, avatarName, avatarThumb }), instead of its usual post-save destination. The wizard creates the agent itself, so the avatar page must not attach the avatar to an agent on this path.
  4. The wizard reads ?avatarId=&avatarName=&avatarThumb= on arrival, attaches the avatar, and moves to step 2.

Entries expire after two hours so an abandoned setup never hijacks a later, unrelated avatar save. A new avatar-producing page joins the flow by calling captureWizardReturn() on boot and returnToWizard() at its save point.

What's next