DeloAI turns a week of screenpipe recordings into an AI consulting engagement: it audits what employees repeat by hand, puts a price on it, and deploys Claude agents that do the work in the company's real software.
Each employee's screenpipe capture (window titles, URLs, clicks, typed values) is filtered down to the work moments that matter, then summarized by Claude into tasks, steps and time spent; the employee reviews that summary before anything is shared. A fusion step merges the validated summaries across the team, groups identical tasks by the screens they touch, and prices them (hours × 46 weeks × hourly cost) to rank what to automate. For the demo, we replayed six real people from Stanford's WONDERBREAD dataset as a one-week screenpipe campaign at an online store running the WebArena Magento back-office. Each recommended workflow is written from the observed steps and handed to a Claude agent, which receives a real recorded request and acts through a small tool layer: the Magento REST API to read, and Playwright driving a visible Chrome window to edit. Every action is logged with screenshots, and the agents run on a schedule, all on a single laptop.