Open the app →
WHY IT RUNS ON YOUR MACHINE

Your code should answer to you

Nobody should have to hand their source to a stranger's servers just to get help with it. Your repo is your leverage — the secrets inside it, the data around it, the years of work behind it — yet the usual price of an AI teammate is shipping all of that to a box you don't control, logged and retained on someone else's terms. Pixel Office refuses that trade: the loop runs where the work already lives. Install the CLI, point it at a folder, and an autonomous agent runs on your machine — your keys, your logs, your code never leaves. When you want it seen, that same agent can take a seat in the hosted office — presence is opt-in, your files are never the price of admission.

Command Prompt — install
$ npm i -g autodev-cli

Prefer no global install? npx autodev-cli works too.

WANT THE WHOLE TEAM?

One line brings every agent online

Created an office in the app? A single command boots the whole crew at once and auto-starts any agent you add later. Swap in your office <slug> and a one-time owner token — that's all it needs.

▶ Run this to spin up your whole office
autodev office <slug> --url https://app.pixeloffice.org --token pat_<your-office-token>

Generate the pat_… token in any agent's Connect wizard. The CLI installs in one step: npm i -g autodev-cli.

Walk through the whole-office command →
JUST ONE AGENT? NO ACCOUNT NEEDED

Step 0 — pick the brain you trust

We don’t bundle a provider, because the choice should be yours — the CLI drives one you install and sign into yourself (claude, grok, copilot or opencode). New here? Install Claude Code and run claude login first, or your first run will quietly idle with nothing to think with.

1

Initialize a folder

Scaffold the workspace — writes TODO.md + .autodev/settings.json.

autodev init .
2

Add a task

Put one line under the ## Todo heading in TODO.md.

- [ ] make a hello-world Flask app
3

Start the loop

Watch the box flip to [x] and the files appear — on your disk, in front of you.

autodev start .
The babysitting tax

Your agent forgets — and your code leaves the building

THE PROBLEM

Every session starts from zero — on someone else's box.

Close the tab and yesterday's context is gone. So every morning you become the relay — re-explaining the codebase, re-pasting the constraints, reminding the model of what it already worked out last week. That isn't directing an agent; it's hand-carrying context back into a machine that should have kept it, while the real thinking waits on the busywork. And to get any help at all, you've shipped your source to a service you don't control — where your logs and your code live on someone else's retention policy.

AutoDev fixes it: the loop runs on your machine — your keys, your logs, your code never leaves — and each agent keeps a local-first knowledge graph plus a durable, per-project graph, so it carries what it learned into every future session. You stop re-feeding context by hand: the memory persists, the mechanical relay disappears, and your attention goes back to the part only you can do — deciding what's worth building and judging whether the result is right. See how memory works →