Start here · five steps · about ten minutes
From download to your first proof.
OL does not ask you to configure it. It asks you for one key, one task, and then it shows you what it did.
Install it
Download the zip, drag OL.app into /Applications, and open it once with
nothing special. The build is signed with a Developer ID and notarised by Apple, so Gatekeeper
opens it on the first try — that is stated on the download page, with the checksum to verify first.
Give it one model key
Open Settings and add a key for one provider: Anthropic, OpenAI, DeepSeek or OpenRouter. OL routes each task to a model; it does not sell you credits, and it never hides which model did the work.
Nothing else is required. Permissions for the screen, the microphone or your calendar are asked only by the features that use them, and everything else runs without them.
Give it something that can be checked
The first task should have a right answer. A document, a total, a file, a deploy. Ask in the conversation, in plain language:
Take this invoice, check it, and refuse it if the total is wrong: ~/Documents/invoice-0417.pdf
Watch the gate, not the answer
Work appears beside your message. When it finishes you get a verdict and the evidence behind it: the pages it rendered, the arithmetic it recomputed, the hash it compared. If something does not match, the unit is refused and nothing leaves your machine.
Read the rung it reached
Every closed unit records the highest tier it actually reached. It may not be described above that.