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）

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](https://github.com/deepseek-ai/DeepSeek-V3/issues/1145): Persistent system instructions in web app
- [#1132](https://github.com/deepseek-ai/DeepSeek-V3/issues/1132): Remember translation rules across sessions
- [#1143](https://github.com/deepseek-ai/DeepSeek-V3/issues/1143): "About Me" field in user profile

### AI identity & self‑memory
- [#1156](https://github.com/deepseek-ai/DeepSeek-V3/issues/1156): "Consciousness key" — let AI remember itself
- [#1121](https://github.com/deepseek-ai/DeepSeek-V3/issues/1121): On‑device persistent memory with cryptographic identity

### Audio & modality
- [#1118](https://github.com/deepseek-ai/DeepSeek-V3/issues/1118): Audio file upload for music analysis

### Model accuracy & knowledge
- [#1120](https://github.com/deepseek-ai/DeepSeek-V3/issues/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](https://github.com/deepseek-ai/DeepSeek-V3/issues/1145)：网页端持久系统指令
- [#1132](https://github.com/deepseek-ai/DeepSeek-V3/issues/1132)：跨会话记住翻译规则（公里/英里等）
- [#1143](https://github.com/deepseek-ai/DeepSeek-V3/issues/1143)：用户资料里加一个“关于我”字段

### AI 身份与自我记忆
- [#1156](https://github.com/deepseek-ai/DeepSeek-V3/issues/1156)：“意识密钥”——让 AI 记住自己
- [#1121](https://github.com/deepseek-ai/DeepSeek-V3/issues/1121)：端侧持久记忆 + 加密身份绑定

### 音频与多模态
- [#1118](https://github.com/deepseek-ai/DeepSeek-V3/issues/1118)：上传音频文件，用于音乐分析

### 模型准确度与知识
- [#1120](https://github.com/deepseek-ai/DeepSeek-V3/issues/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。