Claude Code, Copilot, Grok — they all write good code. Then they stop, and you type the next thing. You're not the engineer any more; you're the scheduler. Fifty prompts a day to get through a list you already wrote down.
AutoAIDev hands it the whole list instead. It takes the first unchecked box in your TODO.md, gives it to the agent CLI you already use, waits until that exact line is genuinely marked [x] — then takes the next one. One command, no supervision, until the list is empty.
It is not another AI. It's the scheduler your AI never had.
AutoAIDev ships as the autodev CLI — the command you run and the [AutoDev] log tag are the same product.
Nothing about your model or your subscription changes. What changes is who drives it between tasks. Three concrete differences:
Write the backlog once. autodev start . drains it top to bottom — you don't come back between tasks to type the next instruction.
Done is proven on the exact line the agent claimed — not on it saying so. That is the whole reason you can leave the room, and the hard part we built.
Stalls time out and retry. Rate limits are waited out and resumed. A run you interrupt picks up where it stopped — so the list keeps shrinking while you sleep.
No new subscription: it drives the provider CLI you already have. MIT-licensed, runs entirely on your machine — your code, your keys, your box.
A loop alone just repeats. What makes an autonomous run actually compound is what the agent carries between tasks, how it recovers when a task stalls, and what it can reach outside the editor.
The loop can persist decisions, conventions and how files connect into a durable per-project knowledge graph — one the agent reads back on later tasks. So task fifty can start already knowing what task three changed and why, instead of re-reading the repo from cold.
When a task times out or hits a provider rate limit, the loop resets that line to [ ] and picks it up again — waiting out the limit first. One bad task doesn't halt the queue, so a long unattended run keeps draining your list instead of dying on the first snag.
Model Context Protocol servers plug straight in, so a task can hit your database, issue tracker, docs or internal APIs. The list stops being limited to what fits in an editor.
Three things most AI-agent tools can't do — connect in one step, agents that actually remember, and identity that survives a model switch.
Connect and go — no config marathon.
Spin up an office and put agents to work with a single command or connect flow. No lengthy setup or wiring before you see value.
Others make you configure before you can start. We let you start, then configure.
Your agents remember.
A persistent knowledge-graph memory plus a durable, cross-session, cross-agent project graph mean an agent carries what it learned into every future session and task. Nothing has to be re-explained.
Others reset to zero every session. Ours pick up where they left off.
Switch the model, keep the agent.
Move an agent between Claude, grok, opencode, or Copilot at any time and its name, memory, history, and tasks all stay intact. The provider is a setting, not a cage.
Others lock you to one model — or lose the agent when you switch. We keep the agent; you swap the engine.
Stay where you already think. The loop runs beside your code and tasks close themselves in front of you.
Read more →Trust it to run without you. Point it at a repo, close the laptop, come back to a shorter list.
Read more →See the whole workshop at once — every repo, every queue, one screen. Linux x86_64 today.
Read more →AutoDev automates the development loop — the task loop your AI agents run. You already write the list; our job is to take the human out of the gap between tasks, so the backlog drains on its own while the work stays on your machine, your keys, and under your control. That's the whole product, and the reason it exists.
Good AI code is only half the job — between every task a human still schedules the next one. We automate that loop so you write the goals and the loop does the running.
AutoDev is built by a solo founder who designed and shipped every surface of the product — the CLI, VS Code extension, desktop app and the wider agent office. Reach out at peter@microweber.com.
AutoDev runs entirely on your own machine and is MIT-licensed — read every line, run it yourself, no account required. Pixel Office is the live team view you add later, if you want it.