DeepYard
A

Agent Memory Distillation

Training-free knowledge transfer from large to small agents via hierarchical memory

Open SourceFree

About

Research framework enabling efficient agent distillation without fine-tuning. Transfers structured knowledge from large teacher agents to smaller student agents through three memory layers: workflow (task decomposition), semantic (domain knowledge), and episodic (execution trajectories). Allows resource-constrained deployments to leverage insights from powerful agents while maintaining performance.

Details

Type
Integrations
Language

Tags

researchmemorymulti-agentautonomousopen-sourceframework