{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/a7ce4779aece4c67bf851dd4a2560885\" frameborder=\"0\" width=\"2026\" height=\"1519\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1519,"width":2026,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1519,"thumbnail_width":2026,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/a7ce4779aece4c67bf851dd4a2560885-dcc281ebf8decf4a.gif","duration":221.943,"title":"Cerebell - Active Memory for AI","description":"Cerebell Demo - Active Memory in Action\n\n1,000 records of venture capital operational data. Same LLM, same retrieval stack, side-by-side. The left panel runs a conventional memory pipeline. The right runs Cerebell enabled.\n\nFour queries show the difference: an LP commitment with conflicting historical values, a board seat with a full supersession chain, a cross-domain synthesis across valuation, leadership, and compliance, and a live ingestion where a new fact resolves against the entire memory state in real time.\n\nConventional memory surfaces stale records alongside current ones and can't distinguish between them. Cerebell resolves state before the LLM ever sees context. The result is structurally coherent memory that compounds with scale and time.\n\n3:41 · Built and narrated by Ezra Brezina, founder of Cerebell."}