# China just made AI memory startups sweat 💀 Tencent open-sourced Agent Memory and it runs 100% locally on SQLite — retained record

> Retained reddit record from reddit:AI_Agents.

Canonical: https://highsignal.app/data/records/e1701bd08720ea55

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:AI_Agents)  published 2026-10-13 06:32 ingested 2026-10-14 02:36

[original source ↗](https://www.reddit.com/r/AI_Agents/comments/1x4pdtc/china_just_made_ai_memory_startups_sweat_tencent/)

## retained text



```
Saw that Tencent open-sourced Agent Memory, a long-term memory layer for AI agents. A few things stood out to me:

- Runs 100% locally on SQLite. No cloud APIs, no separate vector database to host
- They report about 61% fewer tokens
- PersonaMem accuracy went from 48% to 76%
- Memory is readable and traceable, so the agent can point back to the original conversation logs instead of relying on a black box of embeddings

The traceability part is what I find most interesting. Most memory setups I've tried just compress everything into vectors, and when the agent "remembers" something wrong you have no idea why.

If this holds up at scale, I wonder what it means for the hosted memory startups. Has anyone here tried it yet? Curious how it handles memory growing over months of use.
```



document fetched 2026-10-14 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1x4pdtc/china_just_made_ai_memory_startups_sweat_tencent
