How to set up Dots in ChatGPT: a step-by-step guide
Making a Dot takes about ten minutes, and deciding what it may do without asking deserves far more thought than choosing its colour.
Five agents, one sensible order: secure access, connect the minimum, write the rules down, and make the first tasks ones that cannot do harm.
Much of the trouble people report in their first week with an agent comes from doing things in the wrong order: connecting everything, then writing rules after something has already gone out. Each guide here runs the other way round. You confirm that your plan and country qualify, grant as little access as possible, write down what the agent may do alone, and only then hand it a task that reads rather than acts. Setup takes from about five minutes for Muse to half an hour for OpenClaw.
Making a Dot takes about ten minutes, and deciding what it may do without asking deserves far more thought than choosing its colour.
Muse is up and running within minutes, which is exactly why your boundaries message should go out before its first errand does.
Spark arrives already knowing your Google account, so setting it up is mostly a matter of subtraction: fewer apps, fewer tasks, closer watching.
Claude has no agent switch to find: on Pro and Max it judges for itself when a message calls for a longer job, so setup is about what it can reach.
Allow about half an hour to go from a bare machine to a self-hosted agent that answers in Telegram and follows rules you wrote yourself.