[arXiv]score: 0.49
Federation over Text Shares Metacognitive Insights Across Multi-Agent Systems Without Data Sharing
May 26, 2026
FoT enables multiple LLM agents solving different tasks to iteratively federate reasoning traces to a central server that distills cross-task insights, improving reasoning without sharing problem instances or gradients.
cs.LGcs.AI
HOW THIS AFFECTS YOU
●
builderYou can use FoT's insight-sharing pattern to improve reasoning across isolated agent deployments that can't share raw data, relevant for enterprise multi-tenant AI systems.
●
researcherFoT's semantic-level federation without gradient optimization is a novel alternative to federated learning for privacy-preserving multi-agent reasoning improvement.