The AI coding agent you supervise
Ship real code with an AI agent —
and stay in control.
Describe a change in plain English. AI OS plans the work, writes the code, and runs the tests — pausing for your approval at every step that matters. It commits nothing without your yes.
macOS · Python · Open source · MIT
A full task, start to finish — plan → approve → code & test → approve → commit. (~2 min, with sound)
How it works
Three checkpoints. You decide at each one.
AI OS runs a fixed pipeline and stops at three gates where a human has to say yes. Between the gates it works on its own; at the gates, it waits for you.
Review the plan
It reads the relevant files and proposes a step-by-step plan — exactly what it will do and which files it will touch. Nothing is edited until you approve. Reject it and it starts over with your feedback.
Review the code
It writes the changes and runs your test suite. If tests fail after a few retries, it stops and shows you the failures instead of plowing ahead. You see the real diff before anything is kept.
Review the commit
Read the diff, edit the commit message, and choose: commit straight to your branch, or open a pull request for review. Nothing reaches your repository without your sign-off.
Why it's different
Built to do real work without going rogue.
You stay in control
Three human checkpoints — plan, code, commit. Each is a real pause, not a notification you can miss.
Cheap by default
Routes through free AI providers and rotates between them. A hard spending cap stops surprise bills before any call is made.
Real software work
It doesn't just suggest — it edits files, runs the tests, reviews its own diff, and commits. End to end.
Runs on your machine
Your code, your keys, your computer. Nothing is uploaded to a service you don't control.
Remembers your project
Keeps a searchable memory of past work and decisions, so you repeat yourself less over time.
Use it your way
A web dashboard, a chat page, the command line, or Slack and Telegram — same engine behind each.
Get started
Running in about five minutes.
AI OS runs on your own Mac. You bring at least one AI provider account — a free one is fine. No data leaves your machine except the calls you choose to make.
Clone the repo and run setup
Creates a virtual environment, installs dependencies, and copies the example config.
# grab the code git clone https://github.com/philipposk/AI-OS.git cd AI-OS/ai_company # one-shot setup ./infrastructure/deploy.sh
Add one provider key
Open .env and paste in at least one key. A free tier (OpenRouter, Groq, NVIDIA NIM, or local Ollama) is enough to start. Paid Anthropic is optional.
# .env — any one of these works OPENROUTER_API_KEY=... GROQ_API_KEY=... MAX_COST_PER_RUN=0.50 # hard spending cap
Open the dashboard
Launch the web dashboard, type your first task, and approve your way through the three checkpoints. Prefer the terminal? python cli.py does the same thing.
streamlit run ui/dashboard.py
Hand it a task. Keep the final word.
Open source, free to run, and built so an AI can do the heavy lifting while you stay the one who ships.