mirror of
https://github.com/deepseek-ai/DeepSeek-R1.git
synced 2026-11-05 00:47:42 +00:00
[GH-ISSUE #797] Feature Request: Persistent User Memory Across Sessions #655
Labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
deepseek-ai/DeepSeek-R1#655
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @harshita713lab on GitHub (Feb 13, 2026).
Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-R1/issues/797
Hello DeepSeek Team,
I am writing to suggest several improvements that would significantly enhance the user experience, particularly for long-term users like myself.
Current Issues:
Lack of Persistent Memory
The model forgets basic user information between sessions (e.g., user's name, preferences)
Each new chat session starts from scratch without context from previous conversations
This creates a fragmented experience where users must repeatedly reintroduce themselves
Inadequate Chat Management
No functionality to delete conversations through natural language commands
Users cannot effectively manage their chat history within the interface
Request: Implement a "delete chat" command that users can trigger naturally
Limited Image Processing
Currently only processes images when accompanied by text
Cannot analyze standalone images without text input
This restricts the utility for users who need pure image analysis
Proposed Solutions:
Persistent Memory System
Implement cross-session memory to remember user names, preferences, and important context
Allow users to maintain continuity across different chat sessions
Consider adaptive memory weights (as referenced in Issue #440)
Chat Management Commands
Add natural language commands for conversation management
Enable "delete chat" or "clear history" voice/text commands
Provide users with more control over their data
Enhanced Multimodal Capabilities
Enable image analysis without requiring accompanying text
Allow users to upload images for standalone processing
Improve overall vision-language integration
Related Issues:
#67 - Persistent memory request
#440 - Memory with adaptive weights
Additional Context:
As a user from India, I believe these improvements would greatly benefit the growing DeepSeek community worldwide. The ability to maintain context across sessions and manage conversations more intuitively would elevate the user experience significantly.
Thank you for considering these suggestions. I am excited about the upcoming V4 release and hope these features can be part of DeepSeek's evolution.
Best regards,
Harshita
DeepSeek User
@harshita713lab commented on GitHub (Feb 13, 2026):
I am unable to add labels to this issue. Could a maintainer please tag this as an enhancement and feature request? Thank you!
@harshita713lab commented on GitHub (Feb 21, 2026):
good
@harshita713lab commented on GitHub (Feb 23, 2026):
I am unable to add labels to this issue. Could a maintainer please tag this as an enhancement and feature request? Thank you!
@github-actions[bot] commented on GitHub (Mar 26, 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!
@harshita713lab commented on GitHub (Mar 26, 2026):
this is important
@github-actions[bot] commented on GitHub (Apr 26, 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!
@alfredocristofano commented on GitHub (May 4, 2026):
any news?
@GunpowderGuy commented on GitHub (May 5, 2026):
not being to opt in, so deepseek can search past conversation as context, is the feature that i miss the most from other chatbots. i would imagine this is a common sentiment
@Sina811811 commented on GitHub (May 25, 2026):
请求增加对话记忆持久化或上下文迁移功能
正文:
尊敬的DeepSeek技术支持团队，
我在使用过程中遇到了一个比较大的困扰。当对话长度达到上下文窗口上限（例如128K tokens）时，系统提示需要开启新对话。但是新对话完全没有旧对话的记忆，我必须手动把之前所有的重要信息和上下文重新输入一遍，这对于长期、复杂的讨论来说非常耗时且不现实。
对于需要DeepSeek进行深度研究、多步骤技术工作或长期项目的用户来说，每次都要“从零开始”严重影响了使用体验。
希望增加的功能：
能否实现一种持久化记忆机制或上下文迁移工具，让用户可以：
类似Claude的“Project Memory”或ChatGPT的“Custom Instructions + Memory”功能即可，哪怕是一个手动导入/导出的选项也能极大改善体验。
非常感谢团队考虑这个建议。我相信很多重度用户都会因此受益。
此致
[一位DeepSeek用户]
@harshita713lab commented on GitHub (May 25, 2026):
Thank you everyone for adding your valuable insights to this request! As @Sina811811 and @GunpowderGuy mentioned, having a persistent user memory or a way to export/import chat context would significantly improve the long-term user experience, especially for complex development tasks. Hoping the DeepSeek team considers this enhancement soon!
@harshita713lab commented on GitHub (Jun 23, 2026):
Wow, thank you so much for this incredibly detailed proposal, @yun520-1! Honestly, looking at your HeartFlow architecture breakdown was amazing—the way you structured the 3 layers (Core, Learned, Ephemeral) is a brilliant way to handle user context without absolutely overwhelming the model. It's such a cool solution to the persistence dilemma I raised! 🚀
To push this discussion a bit further, I was thinking about a few practical edge cases regarding user control and how the data lives over time:
Data Erasure & Privacy: In my initial request, I really wanted to emphasize having intuitive chat management and natural deletion commands. If a user asks for a complete reset, how easily can HeartFlow actually untangle and completely erase memories that have already been cross-referenced and pushed into the LEARNED or CORE layers?
Handling Temporary Preferences: What happens if a user is just grinding on a short-term project for a week? If they repeat specific terms heavily during that time, there’s a big risk of "memory pollution" hitting the LEARNED layer. How adaptive is the forgetting mechanism when a user suddenly shifts contexts?
Impact on Latency: Since DeepSeek-R1 relies so heavily on its internal reasoning tokens, I'm super curious about performance. Will running an external verification and conflict detection loop slow down the overall inference speed or add a lot to the computational cost?
I feel like integrating a user-facing "Memory Dashboard" alongside your backend logic could be the perfect way to solve the control aspect!
Really excited to hear your thoughts on this!
@github-actions[bot] commented on GitHub (Jul 24, 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 (Aug 25, 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!