The AI operating system in your browser
Hand Aria a goal and watch a team of agents plan, research, write & run real code, and design it — live. With a real language model running entirely on your machine.
Not a chat box. An operating system.
Every capability is real — it works the moment you open it, and gets sharper when you plug in a model.
A real multi-agent team
Seven specialists, orchestrated live. Atlas plans, the team builds in parallel, Echo reviews — every handoff streams in real time.
A brain in your browser
Download a small Llama or Qwen and run it fully on your machine via WebGPU. No server, no key, completely private.
Live code execution
Real Python (CPython on WASM) and sandboxed JavaScript run in-browser. Agents don't just write code — they run it.
Real tools, real effects
Live web search with cited sources, real downloadable artifacts, and image generation. Not scripted text — actual work.
Mission-control graph
A live graph of agents reasoning, calling tools, and handing off — with one-click mission replay.
Talk to it
A hands-free voice mode that listens, transcribes, and speaks back. Plus a full window manager, dock, and Spotlight.
Watch the whole team work
Atlas routes the plan, specialists run in parallel, Echo reviews — nodes light up by status, edges flow on handoff, and each agent shows the tool it's using.
Agents that run real code
Forge writes Python and JavaScript and executes it in the browser — CPython compiled to WebAssembly, JS in a sandboxed worker. Real output, real artifacts.
Grounded in real sources
Sage calls a web-search tool mid-mission and grounds findings in live, cited results — so the work is real, not hallucinated.
Speak, and it answers
A JARVIS-style voice mode listens continuously, transcribes you live, and speaks back — natural, hands-free conversation with your agents.
Three brains, one interface
The agents are brain-agnostic — written once, they run on whichever source you pick.
Simulated
Deterministic, offline, zero-cost. Works the instant you open it — and still uses the real tools.
Local (WebGPU)
A real Llama / Qwen running entirely in your browser. No server, no key, fully private.
Your API key
Bring your own OpenAI or Anthropic key for maximum capability. The key never leaves your browser.
Meet the team
Next.js · React · TypeScript · Tailwind · Zustand · WebLLM (WebGPU) · Pyodide (WASM) · the Web Speech API — typed end-to-end, unit-tested, CI on every push.
See it for yourself
It boots in seconds and works with zero setup. Give it a mission and watch the team go.