OpenAI Dots: Rebuild Its Custom Rules as a Python Gate

OpenAI shipped dots on September 29: always-on agents that live in ChatGPT, get their own cloud computer and browser, and connect to more than 4,000 apps. Within a week, Forbes, DataCamp and a long tail of tech blogs were explaining them. The part that deserves your attention is not the always-on bit. It is the permission model. Every action a dot takes falls into one of four buckets, you set the bucket per action, and an auto-review step sits in front of anything that touches an account.

There is one catch for developers. Per DataCamp's rundown, dots have no public API yet; you set them up in the ChatGPT desktop app or on the web. So if you want that behavior inside your own product, with your own tools and your own audit trail, you have to build it. That is what this guide does: a small, testable Custom Rules gate in Python that sits between gpt-6-astra (the model dots run on) and your tools, plus an 'away mode' that mirrors how dots behave when you are not at your desk.