Getting started
OIOXO opens at oioxo.com. There is nothing to install and no account to make. The whole editor — and the AI — runs in your browser tab. The first screen (the hearth) offers Continue where you left off, then New · Open · Import · Clone, then the three steps below with a runnable example. Want the long version, with every button named? Your first hour walks the whole loop.
1. Pick a model
The model sheet has three modes: On device, Auto and Your key. On device downloads a model once — OIOXO Light is about 540 MB, OIOXO Prime about 1.2 GB, and an open model such as Qwen2.5 Coder 1.5B is just under 1 GB — and runs it on your GPU or CPU. Every row shows speed, size and context, and nothing downloads without asking. Details in Models and Modes.
2. Describe the change
Type what you want — a pomodoro timer I can use — and press Enter. The first time you send a message OIOXO asks where to work:
- Open a folder on this device (Chrome and Edge). The folder is opened in place; nothing is uploaded.
- Import a zip, or Clone a Git URL.
- Just try it in a private folder inside the browser, called
my-project.
The same gate asks for a trust preset: Safe confirms every action the agent takes; Balanced auto-approves reads and edits and still confirms shell commands and deletions; Autonomous auto-approves everything — use it only in a folder under version control. Change it later in Settings → Agent.
3. Review what landed
In Agent mode the model writes files, and each write asks for your consent (once, for the session, or always — per the preset). The Changes panel lists every pending edit with Keep / Undo and Keep all / Undo all, and every agent run is a checkpoint you can revert. Preview opens the result in a built-in browser, and the receipt says whether the page ran clean or what still needs fixing.
Be concrete about what you want to see: “a snake game with a score counter and a restart button” builds better than “a game”. A noun phrase for a thing you want is treated as an order to build it — the first action is to write the file.
What to expect from the on-device tier
A first message on a laptop without a GPU can take a few minutes while the model reads the project — the strip shows what it is doing (reading your request · ~78s, then writing…). Later turns are much faster because the prompt is cached. Small models build simple pages well and larger apps poorly. When a run does not produce a working file, OIOXO offers a bigger model you already have, or one of your keys — it offers, and never re-sends your request on its own.
Come back later
History lists your chats per folder with rename, delete and export. Reloading restores your tabs. Settings persist on this device; Settings → Your settings on another device exports them (your keys stay behind, on purpose — see Your own keys).
On a phone
Open oioxo.com in Chrome or Safari and add it to your home screen. The phone is a companion: watch a run, send a prompt, make light edits. It is not the full IDE — see System requirements for what each phone can run.