Long-term collaborative AI agents often face an awkward "three-week trajectory": the first week is impressive, the second week causes unease due to frequent forgetting and drifting understanding, and by the third week, they are ultimately reduced to ordinary instant query tools. To completely solve the pain points of fragmented memory and chaotic time sequences in long-term collaborative tasks, Tencent Hunyuan officially launched today (May 28, 2026) a smart memory plugin specifically designed for long-term collaborative agents like Openclaw — Hy-Memory.

Tencent Hunyuan

As the "second brain" of an agent, Hy-Memory beats mainstream frameworks on authoritative public test sets LongMemEval and PersonaMem, thanks to its three core technologies. Evaluation data shows that Hy-Memory can reduce memory quantity by over 70%, and increase information density per memory by over 45%; when processing long contexts, token consumption decreases by 35%, and memory update speed increases by 20%.

QQ20260529-090053.jpg

Three Core Technologies, Reshaping AI Memory Mechanism

  • Layer One: Six-Level Memory Framework (Precise Placement)

    Hy-Memory refuses to dump all information into the vector database at once. Instead, it decouples memory into six levels: L1 raw traces, L2 atomic facts, L3 identity profiles, L4 session summaries, L5 mental models, and L6 prospective intentions. It precisely calls the corresponding level based on user questions, streamlining prompts and avoiding irrelevant text diluting the model's attention.

  • Layer Two: System1/System2 Dual Systems (Balancing Speed and Depth)

    Memory processing is split into two independent channels: System1 (day shift mode) extracts facts and updates summaries (L1-L4) in real-time within one second after the user presses enter, ensuring immediate use in the next conversation; System2 (night shift mode) runs asynchronously in the background, using seconds to minutes to deeply consolidate users' mental models and knowledge networks (L5-L6), making the agent smarter over time without blocking the main process.

  • Layer Three: Evolutionary Chain Mechanism (Preserving Causal Traces)

    This is Hy-Memory's secret weapon. When users' views or habits change, traditional systems either "overwrite old with new," losing historical experiences, or "pile up disorderly," leading to incorrect recall. Hy-Memory connects new and old memories into an "evolutionary chain" through the supersedes pointer. Once any node in the chain is triggered, the entire evolutionary path automatically unfolds. The agent not only remembers the user's latest conclusion but also revisits the full evolution of their attitude, thus avoiding past "pitfalls."

  • QQ20260529-090102.jpg

Extreme Performance and 5-Minute Rapid Deployment

In terms of performance, Hy-Memory achieves ultra-fast write speeds comparable to mem0 (8 times faster than Graphiti), with only one-third the number of memories compared to mem0. It uses local embedded storage by default, and no Docker deployment, external services, or Qdrant database are required, with data automatically persisted locally.

To adapt to different hardware and scenario needs, Hy-Memory offers Lite (Lightweight), Pro (Professional), and Ultra (Ultimate) configurations. These three tiers share the same SDK, and switching between them is seamless simply by changing switches. The project has now officially launched, and users can obtain more guides through its official website or user documentation