Hy Memory 2.0
Tencent's open-source agent memory stack: a 30B-A3B memory-writer model, an SDK, and the MemHub local memory app.
Overview
Tencent open-sourced Hy Memory 2.0 on October 16, 2026, covering three parts: Hy-Memory-30B-A3B, a mixture-of-experts memory-writer model with about 3B active parameters that decides what an agent stores and how new facts overwrite old ones; the Hy Memory SDK for chat, work, and code memories; and MemHub, a local memory app with a CLI and a Mac app that imports history from supported AI tools into one local store. The SDK and MemHub are on GitHub under Tencent-Hunyuan and the weights are on Hugging Face and ModelScope, with the method described in arXiv 2610.22196. Tencent reports a 55.49 macro average versus 54.62 for Claude Opus 5 on 20% benchmark subsamples; coverage notes the margin is under 0.6 points on three of four frameworks and is unverified.
This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.
Key features
- Hy-Memory-30B-A3B memory-writer model (30B MoE, about 3B active); Apache-2.0 reported by secondary coverage
- Hy Memory SDK for chat, work, and code memories
- MemHub: local memory app with a CLI and Mac app that imports history from supported AI tools
- Weights on Hugging Face and ModelScope; code on GitHub (Tencent-Hunyuan)
- Vendor-reported 55.49 macro average vs 54.62 for Claude Opus 5 on subsampled benchmarks
Use cases
- Adding persistent, updatable memory to coding and chat agents
- Keeping one local memory store across several AI tools
Pricing
Free and open source; self-hosted.
Pricing reflects the source record and may change. Confirm current terms with the provider.
Dates and sources
- Released
- 16 Oct 2026
- Record updated
- 18 Oct 2026
- Record last checked
- 18 Oct 2026