Skip to content
OIOXO

Airplane mode

Most “AI IDEs” are a thin window onto someone else’s datacenter — cut the cable and they’re a text editor. OIOXO is built the other way around: the model, the runtime, and your project live on your device, so losing the connection loses you very little.

What runs on your device

  • The AI model. Downloaded once from a public CDN, then stored on your disk (browser origin storage). Generation, planning, and repair run locally — online or not.
  • The runtimes. The WebAssembly/WebGPU inference engine, the code runners, and the preview environment execute in your browser.
  • Your files. Projects, snapshots, settings, skills, and history are stored by your browser on your machine. Saving, editing, and searching never need a server.
  • Git. Local history and commits work offline; pushing to GitHub obviously resumes when you reconnect.
  • The IDE itself. After your first visit, a service worker keeps the app shell cached, so oioxo.com/oioxo loads with no internet at all.

What the server is actually for

Our servers never see your code, prompts, or output (privacy policy). They do exactly two jobs: accounts (sign-in, Pro entitlements) and fair-use metering (counting tokens against the free allowance — counts only, never content).

Honest limits — offline-tolerant, not offline-forever

Being straight about what needs a connection:

  • The first-ever model download. Roughly a gigabyte from a public CDN. After that it lives on your disk and never re-downloads.
  • Sign-in and Pro unlock. Authenticating and fetching the Pro model key require the server. Already signed in? A cached entitlement keeps working offline for about a day.
  • Usage metering. The free allowance is enforced server-side, so metering needs a periodic connection. Offline, coding continues against your last-known remaining allowance for a bounded grace window (about 24 hours); tokens you use are queued on your device and reported when you reconnect. Past the window — or past your allowance — you’ll be asked to reconnect to continue. No hidden free tier, no fake meter.
  • BYOK cloud models and web search. Calls to OpenAI/Anthropic/etc. and live web lookups are network features by definition.

Try it yourself

  1. Open the IDE and start a project, letting the model finish its one-time download.
  2. Turn on airplane mode (or flip your browser’s DevTools network panel to “Offline”).
  3. Reload the page. The IDE comes back, your project is intact, and the model keeps generating — no internet involved.
  4. Reconnect whenever. Queued usage syncs up quietly in the background.

That’s the whole pitch: the badge above isn’t a marketing widget, it’s the browser telling you the truth.