Tencent Open-Sources Hy Memory 2.0 Agent Memory Framework
Tencent has released and open-sourced Hy Memory 2.0, a comprehensive agent memory framework consisting of specialized models, a memory framework, and the MemHub local management application. The framework introduces specialized layers for chat, work, and code memory tasks, outperforming several existing open-source benchmarks. This release represents a significant advancement in AI memory management, providing developers with robust tools to help AI agents maintain context and continuity across long tasks and multiple tools. Its specialized Hy-Memory-30B-A3B model even scored higher than Claude Opus 5 on specific benchmark tests, signaling strong competitiveness in agentic workflows. The framework features the Hy-Memory-30B-A3B model with approximately 3B active parameters, trained across multiple memory frameworks. Additionally, the MemHub application supports integration with ten popular AI tools and coding assistants via plugins and SDKs.
## BACKGROUND
AI agents typically struggle to retain long-term context, user preferences, and historical decisions across multiple sessions without structured memory layers. Frameworks like Hy Memory aim to solve this by organizing historical interactions into distinct memory types—such as episodic, semantic, and working memory—allowing agents to retrieve relevant information efficiently.