User requests related to memory / persistence (from DeepSeek-V3 issues)用户关于“记忆/持久化”相关的需求汇总（来自 DeepSeek-V3 issues）

[qingkong66]

I've been following the DeepSeek-V3 issue space and noticed a number of requests that all relate to memory, persistence, and user preferences. Since Engram is DeepSeek's core memory technology, I'm collecting them here in case they're useful for the team.

---

Persistent user rules & preferences

• #1145: Persistent system instructions in web app
• #1132: Remember translation rules across sessions
• #1143: "About Me" field in user profile

AI identity & self‑memory

• #1156: "Consciousness key" — let AI remember itself
• #1121: On‑device persistent memory with cryptographic identity

Audio & modality

• #1118: Audio file upload for music analysis

Model accuracy & knowledge

• #1120: Fine‑tune on official Dota 2 Lua API docs

These requests come from different users—researchers, translators, developers, musicians—but they all point to the same underlying need: users want AI to remember how they use it, not just what they say.

I'm posting this here because Engram seems like the natural place for these conversations to land. If any of this is useful, great. If not, feel free to close.

我一直在关注 DeepSeek-V3 的 issue 区，注意到有不少请求都和“记忆”“持久化”“用户偏好”相关。既然 Engram 是 DeepSeek 的核心记忆技术，我把它们汇总到这里，也许对团队有帮助。

持久化的用户规则与偏好

• #1145：网页端持久系统指令
• #1132：跨会话记住翻译规则（公里/英里等）
• #1143：用户资料里加一个“关于我”字段

AI 身份与自我记忆

• #1156：“意识密钥”——让 AI 记住自己
• #1121：端侧持久记忆 + 加密身份绑定

音频与多模态

• #1118：上传音频文件，用于音乐分析

模型准确度与知识

• #1120：用官方 Dota 2 Lua API 文档微调

这些请求来自不同的用户——研究者、翻译者、开发者、音乐人——但都指向同一个核心需求：用户希望 AI 记住“怎么用”，不只是“说了什么”。

我把它们汇总到这里，因为 Engram 是这些讨论最自然的落脚点。如果有用，很好；如果不需要，直接关闭即可。