Skip to content

Repository files navigation

Live Agents

A beautiful workspace to design, configure, and watch your AI agents collaborate — in chat, in tasks, or in a living 3D office when WebGPU is available.

Live demo: csherman-lab.github.io/live-agents

Features

  • Command Center — Write a brief, pick templates, chat with your lead agent, and preview the workspace
  • Text-first workflow — Run agents without 3D; WebGPU is optional for the immersive office
  • Go Live — Enter a 3D office where agents walk, sit, work, and collaborate (Chrome, Edge, Safari 18+)
  • Multi-provider AI — Bring your own key for Google Gemini, OpenAI, or Anthropic Claude (stored locally)
  • Team Designer — Visual node-based editor to create multi-agent teams
  • Provider warnings — Alerts when your team output type (image/music/video) doesn't match your provider
  • Export — Deliverables, HTML report, and full project bundle JSON

Getting Started

npm install
npm run dev

Open http://localhost:3000

API keys

Provider Key URL Capabilities
Google Gemini aistudio.google.com Text, image, music, video
OpenAI platform.openai.com Text only
Anthropic console.anthropic.com Text only

Use Explore demo to browse the UI without a key. Agents only run after you add a key. For image/music/video teams, use Gemini.

Copy .env.example to .env for optional local dev variables (never commit real keys).

Demo guide (showing others)

  1. Open the live demo in Chrome or Safari (not an embedded IDE browser).
  2. Skip the intro, open Settings, and paste a Gemini API key (easiest full demo) or OpenAI/Anthropic for text-only teams.
  3. Pick a Text output team (e.g. Creative Agency) in Teams if using OpenAI or Anthropic.
  4. Write a short brief in the Command Center, then Go Live.

What to tell viewers

  • Explore demo = UI walkthrough only; agents need an API key to actually run.
  • 3D office needs WebGPU (Chrome, Edge, Safari 18+). Overview chat and planning work without 3D.
  • Keys stay in the browser — nothing is sent to our servers.

Keyboard shortcuts

Key Action
G Go Live / enter workspace
S Open Settings
? Show all shortcuts

Scripts

npm run dev          # Dev server on port 3000
npm run build        # Production build
npm run lint         # TypeScript check
npm run test         # Unit tests (Vitest)
npm run test:e2e     # Smoke tests (Playwright)
npm run check        # lint + test + build
npm run assets:generate  # Rebuild 3D models (requires Blender)

Tech Stack

  • React 19 + TypeScript + Vite
  • Three.js (WebGPU) for 3D simulation
  • React Flow for team visualization
  • Zustand for state management
  • Gemini / OpenAI / Anthropic APIs for agent intelligence

3D assets

Original MIT-licensed workspace and character models in public/models/. See docs/assets/art-direction.md and docs/assets/3d-asset-spec.md. Regenerate with npm run assets:generate (Blender required).

License

MIT — see LICENSE.

About

A beautiful 3D workspace to design, configure, and watch your AI agents collaborate in real time

Resources

Stars

Watchers

Forks

Releases

Packages

Used by

Contributors

Languages