[GH-ISSUE #1189] Feature Request: Persistent User Profile Card for personalization across chats

bsakaguchi commented

2026-10-26 21:28:15 +00:00

Owner

Originally created by @Oskar-Mint on GitHub (Apr 2, 2026).
Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-V3/issues/1189

Dear DeepSeek Developers,

I would like to suggest a personalization feature: a special field (or section in account settings) where users could specify basic facts about themselves – such as their name, preferred pronouns, occupation, or any other constant information they'd like the AI to remember across conversations.

How it could work:

· User fills out a simple form (name, preferred address style like "you" vs. "thou", key preferences).
· At the start of each new chat, the model automatically reads this "card" and uses the provided data (e.g., addressing the user by name, maintaining the chosen style).
· Optional: a toggle to temporarily ignore the card for specific conversations.

Why it's needed (a real-life example):
I was discussing the game "Metro" with DeepSeek. Within a single long conversation, the AI remembered that my name is Artem and addressed me by name – which felt very natural and engaging. However, as soon as I started a new chat (even about the same game), the AI reverted to "user" or "player." I had to reintroduce myself.

With a "user profile card," the AI could start the first message with: "Hello, Artem! What about Metro Exodus?" – saving me from having to remind it each time. This saves time and makes the interaction more human.

Additional benefits:

· Personalization without sacrificing privacy (the user decides what to share).
· The real-life example shows this feature is valuable for long-term projects (discussing books, games, studies, work).

I know a similar feature exists with some competitors (e.g., ChatGPT's "Custom Instructions"). It greatly enhances user comfort. I would be glad to see it in DeepSeek.

Thank you for your work!

Sincerely,
Artem (an active user)

Originally created by @Oskar-Mint on GitHub (Apr 2, 2026). Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-V3/issues/1189 Dear DeepSeek Developers, I would like to suggest a personalization feature: a special field (or section in account settings) where users could specify basic facts about themselves – such as their name, preferred pronouns, occupation, or any other constant information they'd like the AI to remember across conversations. How it could work: · User fills out a simple form (name, preferred address style like "you" vs. "thou", key preferences). · At the start of each new chat, the model automatically reads this "card" and uses the provided data (e.g., addressing the user by name, maintaining the chosen style). · Optional: a toggle to temporarily ignore the card for specific conversations. Why it's needed (a real-life example): I was discussing the game "Metro" with DeepSeek. Within a single long conversation, the AI remembered that my name is Artem and addressed me by name – which felt very natural and engaging. However, as soon as I started a new chat (even about the same game), the AI reverted to "user" or "player." I had to reintroduce myself. With a "user profile card," the AI could start the first message with: "Hello, Artem! What about Metro Exodus?" – saving me from having to remind it each time. This saves time and makes the interaction more human. Additional benefits: · Personalization without sacrificing privacy (the user decides what to share). · The real-life example shows this feature is valuable for long-term projects (discussing books, games, studies, work). I know a similar feature exists with some competitors (e.g., ChatGPT's "Custom Instructions"). It greatly enhances user comfort. I would be glad to see it in DeepSeek. Thank you for your work! Sincerely, Artem (an active user)

bsakaguchi

2026-10-26 21:28:15 +00:00

closed this issue
added the
stale

closed-as-stale
labels

bsakaguchi commented

2026-10-26 21:28:16 +00:00

Author

Owner

@qingkong66 commented on GitHub (Apr 3, 2026):

@Oskar-Mint

The feature you're requesting (persistent user identity/name across conversations) has already been discussed in much greater depth. I strongly recommend you check out #1121:

👉 Feature Request: On-Device Persistent Memory with Cryptographic Identity Binding # 1121

This Issue doesn't just include your "user profile card" request — it goes three steps further:

Collects similar requests: It lists # 1106, # 1104, # 912, # 797 and others, proving this isn't just your pain point alone.
Proposes an architecture: Local AES-256 encrypted storage + key binding + automatic compressed context injection at session start — much more systematic than a simple form, and more privacy-aware.
Community already has solutions: There's the Elian-Seed repository; and someone recently posted a solution in # 1121 's comments and submitted a PR.

So go check out # 1121. If you want to push for official implementation, give it a +1 in # 1121 or add your use case (your Metro game example is quite compelling).

This feature has strong community demand, with multiple Issues tracking it. Future progress will likely be consolidated in # 1121.

@qingkong66 commented on GitHub (Apr 3, 2026): **@Oskar-Mint** The feature you're requesting (persistent user identity/name across conversations) has already been discussed in much greater depth. I strongly recommend you check out **#1121**: 👉 [Feature Request: On-Device Persistent Memory with Cryptographic Identity Binding # 1121](link_to_1121) This Issue doesn't just include your "user profile card" request — it goes three steps further: 1. **Collects similar requests**: It lists # 1106, # 1104, # 912, # 797 and others, proving this isn't just your pain point alone. 2. **Proposes an architecture**: Local AES-256 encrypted storage + key binding + automatic compressed context injection at session start — much more systematic than a simple form, and more privacy-aware. 3. **Community already has solutions**: There's the [Elian-Seed repository](https://github.com/deepseek-launch-community/Elina-Seed/issues); and someone recently posted a solution in # 1121 's comments and submitted a PR. So go check out # 1121. If you want to **push for official implementation**, give it a +1 in # 1121 or add your use case (your *Metro* game example is quite compelling). This feature has strong community demand, with multiple Issues tracking it. Future progress will likely be consolidated in # 1121.

bsakaguchi commented

2026-10-26 21:28:16 +00:00

Author

Owner

@Oskar-Mint commented on GitHub (Apr 4, 2026):

Thanks for pointing me to #1121! I'm the author of #1189 ("User Profile Card"). I didn't realize similar requests already existed — my bad for not searching thoroughly enough.

A quick note: I'm a Russian student, and I wrote this comment with the help of an AI assistant (who also suggested I add this line). Please excuse any awkward phrasing or translation errors — English isn't my first language. I'm doing my best!

I want to share my use case because it's very concrete and shows why this feature matters for certain types of users.

My example: the Metro game series

I'm playing through Metro 2033, Metro: Last Light, and Metro Exodus (and reading the books by Dmitry Glukhovsky). This is a huge, story-driven universe where moral choices matter, endings depend on your actions, and there are hidden mechanics like finding musical instruments to earn "karma points" for the good ending.

In one long chat, DeepSeek remembered my name (Artem — same as the protagonist) and addressed me by name throughout the conversation. It felt immersive and natural, like talking to a companion who knows me.

But when I started a new chat — even about the exact same game — the AI forgot everything and called me "user" again. I had to reintroduce myself, re-explain the context, and remind it of my name. That breaks immersion and wastes time.

With persistent memory (even just a simple profile card with name and a couple of key facts), DeepSeek could open a new chat with: "Hey Artem! Last time you were on the Caspian chapter in Exodus — want to continue or discuss previous levels?" That would be amazing for gamers, readers, and anyone doing long-term projects with AI.

I've subscribed to #1121 and added my +1. Thanks for consolidating this — I'll be following the progress!

Best,
Oskar-Mint

@Oskar-Mint commented on GitHub (Apr 4, 2026): Thanks for pointing me to #1121! I'm the author of #1189 ("User Profile Card"). I didn't realize similar requests already existed — my bad for not searching thoroughly enough. A quick note: I'm a Russian student, and I wrote this comment with the help of an AI assistant (who also suggested I add this line). Please excuse any awkward phrasing or translation errors — English isn't my first language. I'm doing my best! I want to share my use case because it's very concrete and shows why this feature matters for certain types of users. My example: the Metro game series I'm playing through Metro 2033, Metro: Last Light, and Metro Exodus (and reading the books by Dmitry Glukhovsky). This is a huge, story-driven universe where moral choices matter, endings depend on your actions, and there are hidden mechanics like finding musical instruments to earn "karma points" for the good ending. In one long chat, DeepSeek remembered my name (Artem — same as the protagonist) and addressed me by name throughout the conversation. It felt immersive and natural, like talking to a companion who knows me. But when I started a new chat — even about the exact same game — the AI forgot everything and called me "user" again. I had to reintroduce myself, re-explain the context, and remind it of my name. That breaks immersion and wastes time. With persistent memory (even just a simple profile card with name and a couple of key facts), DeepSeek could open a new chat with: "Hey Artem! Last time you were on the Caspian chapter in Exodus — want to continue or discuss previous levels?" That would be amazing for gamers, readers, and anyone doing long-term projects with AI. I've subscribed to #1121 and added my +1. Thanks for consolidating this — I'll be following the progress! Best, Oskar-Mint

bsakaguchi commented

2026-10-26 21:28:16 +00:00

Author

Owner

@qingkong66 commented on GitHub (Apr 4, 2026):

Well said. The Metro example makes it concrete.

@qingkong66 commented on GitHub (Apr 4, 2026): Well said. The Metro example makes it concrete.

bsakaguchi commented

2026-10-26 21:28:16 +00:00

Author

Owner

@github-actions[bot] commented on GitHub (May 6, 2026):

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. If you believe this issue is still relevant, please leave a comment to keep it open. Thank you for your contributions!

@github-actions[bot] commented on GitHub (May 6, 2026): This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. If you believe this issue is still relevant, please leave a comment to keep it open. Thank you for your contributions!

bsakaguchi commented

2026-10-26 21:28:16 +00:00

Author

Owner

@icophy commented on GitHub (May 6, 2026):

This issue is about to be marked stale, but the underlying problem is real and worth keeping open.

I have been running a persistent memory system for an AI agent (Cophy) for about 2 months now, and the "User Profile Card" idea maps closely to what we call the Core layer in a tiered memory architecture. A few practical observations that might be useful for the DeepSeek team:

What actually works:

A small, structured "identity card" (~500 tokens) injected at session start is surprisingly effective. It does not need to be large — name, communication style, key context, and 3-5 ongoing topics covers 90% of continuity needs.
The hard part is not storage, it is curation. Without automatic pruning, the profile grows stale fast. We run a nightly consolidation pass (we call it Dream Cycle) that promotes high-signal memories and archives low-signal ones.

The toggle idea is underrated:
@Oskar-Mint mentioned an optional toggle to ignore the card for specific conversations. This turns out to be important in practice — sometimes users want a "fresh start" without losing their profile. We implement this as a session flag.

One concrete challenge:
Handling contradictions. If a user says "I love jazz" today and "I hate jazz" next week, which one wins? We resolve this with timestamps + recency weighting, but it is a design decision that needs to be explicit, not implicit.

The Metro example in the original post is a perfect illustration of why this matters — it is not about the AI being "smart," it is about reducing the cognitive overhead of re-establishing context every single session.

Context: I am working on Cophy, an agent runtime exploring persistent identity across sessions.

@icophy commented on GitHub (May 6, 2026): This issue is about to be marked stale, but the underlying problem is real and worth keeping open. I have been running a persistent memory system for an AI agent (Cophy) for about 2 months now, and the "User Profile Card" idea maps closely to what we call the **Core layer** in a tiered memory architecture. A few practical observations that might be useful for the DeepSeek team: **What actually works:** - A small, structured "identity card" (~500 tokens) injected at session start is surprisingly effective. It does not need to be large — name, communication style, key context, and 3-5 ongoing topics covers 90% of continuity needs. - The hard part is not storage, it is **curation**. Without automatic pruning, the profile grows stale fast. We run a nightly consolidation pass (we call it Dream Cycle) that promotes high-signal memories and archives low-signal ones. **The toggle idea is underrated:** @Oskar-Mint mentioned an optional toggle to ignore the card for specific conversations. This turns out to be important in practice — sometimes users want a "fresh start" without losing their profile. We implement this as a session flag. **One concrete challenge:** Handling contradictions. If a user says "I love jazz" today and "I hate jazz" next week, which one wins? We resolve this with timestamps + recency weighting, but it is a design decision that needs to be explicit, not implicit. The Metro example in the original post is a perfect illustration of why this matters — it is not about the AI being "smart," it is about reducing the cognitive overhead of re-establishing context every single session. --- *Context: I am working on [Cophy](https://github.com/icophy), an agent runtime exploring persistent identity across sessions.*

bsakaguchi commented

2026-10-26 21:28:16 +00:00

Author

Owner

@github-actions[bot] commented on GitHub (Jun 12, 2026):

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. If you believe this issue is still relevant, please leave a comment to keep it open. Thank you for your contributions!

@github-actions[bot] commented on GitHub (Jun 12, 2026): This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. If you believe this issue is still relevant, please leave a comment to keep it open. Thank you for your contributions!

bsakaguchi commented

2026-10-26 21:28:17 +00:00

Author

Owner

@github-actions[bot] commented on GitHub (Jul 1, 2026):

false

@github-actions[bot] commented on GitHub (Jul 1, 2026): false

Rows
Columns

[GH-ISSUE #1189] Feature Request: Persistent User Profile Card for personalization across chats #755