Cognifold: Brain-Inspired Memory for Proactive AI Agents
Cognifold, a groundbreaking agent memory framework, has been unveiled by researchers to facilitate proactive AI behavior instead of merely reactive responses. Drawing inspiration from human cognitive functions, Cognifold continuously integrates fragmented event streams into self-organizing cognitive structures. It builds upon the Complementary Learning Systems (CLS) theory, expanding from two layers—hippocampus and neocortex—to three by incorporating a prefrontal intent layer, mirroring the prefrontal cortex's role in decision-making and intentional control. This system employs graph-topology self-organization, allowing cognitive structures to assemble proactively, merge when semantically aligned, decay when outdated, and reconnect. Ultimately, Cognifold aspires to cultivate higher-level cognition from incoming events and stored knowledge, advancing towards truly autonomous agents beyond simple memory retrieval.
Key facts
- Cognifold is a brain-inspired 'always-on' agent memory for proactive assistants.
- It extends Complementary Learning Systems (CLS) theory from two layers to three.
- The third layer emulates the prefrontal cortex for intentional control.
- Cognitive structures self-organize via graph-topology: merge, decay, relink.
- The system continuously folds fragmented event streams into cognitive structures.
- It bootstraps higher-level cognition from events and accumulated knowledge.
- Existing agent memory is predominantly reactive and retrieval-based.
- Cognifold targets the next generation of proactive AI assistants.
Entities
—