MetaCognition

MetaCognition Labs is building a persistent memory layer for AI agents and robots, aimed at developers who need their agents to retain context across sessions instead of starting from zero each conversation. The product targets a real gap in current agent architectures, where retrieval-augmented systems store logs and summaries but lose continuity as histories grow, personalization decays, and long-running tasks drift. This launch matters now because agent deployments are moving from demos to production, and the memory bottleneck is what breaks most of them once real users stick around for weeks or months. The system is modeled loosely on how the human brain organizes experience, with distinct layers the team names after the hippocampus, inferotemporal cortex, and neocortex. The hippocampal layer captures raw events and interactions as they happen and turns them into semantic nodes, the temporal layer organizes sequence and causality across long horizons, and the neocortical layer consolidates repeated signals into durable structure for reasoning and personalization. The company reports leading scores on the LoCoMo (93.3%) and LongMemEval (92.2%) long-term memory benchmarks, with roughly 120ms retrieval latency and no LLM tokens consumed during ingestion. The launch opens a private beta for teams building agents that need to remember users, adapt to changing preferences, and reason over years of interaction rather than a single context window. The broader pitch, echoed in the launch video, is that understanding was never a scale problem and that cognition has to be shaped by lived experience, positioning the company against the pure frontier-model race. Founders are not named publicly on the site, so anyone evaluating should reach out through the beta signup for team details.