Watch it work, live
Every run is a room. Anyone on your team can open it, see the agent's screen and steps, and leave a note it reads mid-task.
Watch it run, step in, and it keeps the lesson.
Get a demoWorks beside the stack you already run
Every run is a room. Anyone on your team can open it, see the agent's screen and steps, and leave a note it reads mid-task.
Refunds, sends and edits wait in a hold window sized to how much you trust the agent on that task. Undo it, correct it, or send it now.
An undo, a correction, a takeover: each one drafts a lesson the agent keeps. Served to any model through the SDK or MCP.
Agents learn your process from the people who own it. Your team drops into a live run, watches, redirects, takes the wheel and hands it back. No screen-share, no shared password, no fix that's forgotten by Monday.
Sits beside your agent instead of inside it, so it works with the model and framework you already chose.
// before a task, ask what the team taught const lessons = await doubleoh.skillsFor(task)
Orders older than 30 days go to a person. Taught by Lina, used 37 times.
"The refund agent tried to pay order #4821 twice. It sat in the hold window, Sam hit Undo, and the agent now checks for a prior refund first."
"Support kept closing tickets that mentioned a chargeback. Lina left one note mid-run. Those tickets go to finance now, every time."
"The ops agent hit a login it had no password for. Our ops lead took the wheel for a minute and handed it back. Nobody shared a credential."
"We let the agent run in practice for a week. By Friday the hold window on simple refunds was down to seconds, because the team had stopped correcting it."
"The browser agent picked the wrong supplier from a dropdown. One correction, and the lesson was there for every agent on the account the next morning."
Examples of what a run catches. Your agents, your rules.