{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/2978fbfe42324e509057ac5fd46b7a70\" frameborder=\"0\" width=\"1890\" height=\"1417\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1417,"width":1890,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1417,"thumbnail_width":1890,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/2978fbfe42324e509057ac5fd46b7a70-37108be11ee154e6.gif","duration":428.313,"title":"Learning and Memory in CopilotKit","description":"This Loom explains how CopilotKit enables learning and memory directly from an application to create smarter, accumulating agent behavior. It demonstrates long-term memory for summarization of spending, including how the order of instructions from memory affected the resulting summary. It then shows learning from in-app usage, such as identifying a $89.99 delta charge and automatically building and applying skills to flag alerts, set a spending alert, and add the transaction details. Finally, it illustrates recording a workflow to approve an over-limit $15,000 AWS charge by filing a policy exception, then reusing the saved skill in a new thread to approve a Google ads charge from anywhere in the app."}