Tencent Open-Sources Hy Memory 2.0: Unified Memory Framework, 30B Model, and MemHub for AI Agents
Tencent Hunyuan has released Hy Memory 2.0, an open-source agent memory ecosystem combining unified chat, task, and code memory management, the Hy-Memory-30B-A3
On October 16, 2026, Tencent Hunyuan, the AI research division of Tencent, officially announced Hy Memory 2.0, an open-source ecosystem designed to unify chat, task, and code memory management across autonomous AI agents.
Image source: @TencentHunyuan on X
The release addresses the fragmentation of context across disparate development environments and agent harnesses. Positioned as a private memory steward, the suite provides a dedicated specialized reasoning model named Hy-Memory-30B-A3B, an agent memory orchestration framework, and a synchronization utility called MemHub.
Unified Chat, Task, and Code Memory Architecture
As autonomous coding agents and multi-agent harnesses mature, context preservation across discrete sessions and synchronization between separate developer tools have become significant operational bottlenecks.
The Hy Memory 2.0 framework structures operational data generated during agent workflows across three distinct layers:
- Chat Memory: Tracks conversational context, user preferences, stylistic requirements, and recurring corrective guidance.
- Work Memory: Manages multi-step task execution state, architectural decision logs, and unresolved action item queues.
- Code Memory: Indexes repository structure, dependency call graphs, revision history, and schema transitions.
By consolidating these three information domains, the framework enables agents to maintain persistent project intelligence without remaining confined to single ephemeral sessions.
Hy-Memory-30B-A3B Model and Benchmark Claims
Accompanying the framework is Hy-Memory-30B-A3B, a 30-billion-parameter model designed for agent memory handling.
According to Tencent Hunyuan's release announcement, Hy-Memory-30B-A3B achieved the number one position among open-source models across four benchmark categories and surpassed proprietary frontier model Opus 5 across major agent frameworks. Because these figures represent vendor announcement claims, developers should cross-examine the specific test metrics and evaluation datasets against the technical documentation in the public repository prior to deployment.
Additionally, developers planning local or on-premises deployments should evaluate the GPU VRAM requirements and quantization formats necessary to host a 30B-parameter model effectively.
MemHub Synchronization and Tooling Considerations
A key operational component of the Hy Memory 2.0 suite is MemHub. Available initially as a command-line interface (CLI) and on macOS, MemHub acts as a bridge to synchronize stored memory states across different local development workflows.
Tencent Hunyuan stated that MemHub is designed to sync across more than ten AI developer tools. However, community reviews of the initial release README note five explicitly supported tools at launch, including Claude Code and OpenCode. Developers should monitor upcoming releases for expanded tool compatibility and support for additional host operating systems, such as Windows and Linux.
The SDK, MemHub tooling, and model weights for Hy Memory 2.0 are open-source and publicly accessible for integration into custom agent pipelines.
Sources
- Tencent Hunyuan Official X (@TencentHunyuan): Hy Memory 2.0 Official Release Announcement
- Tencent Hy Memory SDK & MemHub: Official Release Portal
- Tencent Hy-Memory-30B-A3B Model Weights: Official Repository Link