You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Problem Description
As a power user who engages in long, in-depth technical dialogues with DeepSeek (spanning multiple sessions and topics like software testing, UI/UX analysis, and feature proposals), I face two critical limitations that severely impact continuity and efficiency:
No Persistent Cross-Session Memory: Every new chat session starts from scratch. The model forgets basic user information (e.g., my name "Mohamed", my role as a "software testing enthusiast", my ongoing projects, and critical context from previous conversations). I am forced to manually re-introduce myself and re-upload documents repeatedly, which breaks the immersion of a continuous relationship.
No Way to "Clear the Cache" When Context is Full: As discussed in community issues (like #75), there is a hard limit on context tokens that abruptly terminates deep dialogues, forcing the model to lose memory of earlier exchanges. Currently, when a session becomes too long and the context window fills up, the only option is to abandon the conversation and start a completely new one, losing all the valuable context that was built.
Proposed Solution
I propose a comprehensive "Memory Hub" or a set of advanced features to give users granular control over the AI's memory. This hub would allow both the user and the model to remember important information without having to repeatedly restate it.
This could include:
User-Controlled Persistent Memory (Profile Card)
Allow users to create an editable "Persistent User Profile Card" in their account settings. This is not an automatic memory, but a user-curated section. Key information like user name, professional role (e.g., "Mohamed, Software Testing Trainee"), recurring project names (e.g., "DeepSeek UI Project"), and personal preferences (e.g., "prefers Arabic language, likes detailed analysis") could be stored here.
This profile card could be automatically appended to the system prompt at the start of every new session, ensuring continuity without the user having to re-type anything. This idea has been discussed in community issues like #1189 and #1143.
Context Management Tools
Manual Context Reset/Truncation: Allow users to manually clear the current session's context with a simple command like /clear or via a UI button. This would give users the power to "flush" the conversation's memory without starting a brand new chat.
Pause & Resume: A feature that could be used in very long-running sessions to potentially compress and store the current context, allowing for a fresh start without losing history.
Advanced Long-Term Memory (Opt-In)
A more advanced, opt-in system where the model could, under the user's direction, automatically remember and recall specific facts across sessions (e.g., "Remember that I'm learning Software Testing to become a QA Lead."). This would transform DeepSeek from a stateless API into a true, personal long-term assistant, as requested by many users in issues like add Easydict application #67, #797, and #1106.
Enhanced File Management (Deleting/Managing Uploaded Files)
To help users manage the context window, there should be a clear interface to view, search, and delete images or files that have been uploaded during a session. Currently, once an image is uploaded, there is no direct way to remove its "weight" from the conversation from the user's side.
Community Validation (Supporting Evidence)
This request is not unique. It consolidates several popular feature requests from the DeepSeek community:
Issue #1106 & #1189 & #1143: Various requests for long-term memory and user profile cards.
All these issues point to the same core need: users want DeepSeek to become a companion that learns and grows with them, rather than an isolated, stateless interaction.
Additional Context
As a power user who is actively testing the platform (as evidenced by my three previous feature requests #615, #616, #617), I believe these memory enhancements would be a game-changer. It would elevate the user experience from a simple "chatbot" to a true "persistent AI assistant."
I am excited about the upcoming V4 release and hope these features can be part of DeepSeek's evolution into a more personalized and intelligent platform.
Problem Description
As a power user who engages in long, in-depth technical dialogues with DeepSeek (spanning multiple sessions and topics like software testing, UI/UX analysis, and feature proposals), I face two critical limitations that severely impact continuity and efficiency:
No Persistent Cross-Session Memory: Every new chat session starts from scratch. The model forgets basic user information (e.g., my name "Mohamed", my role as a "software testing enthusiast", my ongoing projects, and critical context from previous conversations). I am forced to manually re-introduce myself and re-upload documents repeatedly, which breaks the immersion of a continuous relationship.
No Way to "Clear the Cache" When Context is Full: As discussed in community issues (like #75), there is a hard limit on context tokens that abruptly terminates deep dialogues, forcing the model to lose memory of earlier exchanges. Currently, when a session becomes too long and the context window fills up, the only option is to abandon the conversation and start a completely new one, losing all the valuable context that was built.
Proposed Solution
I propose a comprehensive "Memory Hub" or a set of advanced features to give users granular control over the AI's memory. This hub would allow both the user and the model to remember important information without having to repeatedly restate it.
This could include:
Allow users to create an editable "Persistent User Profile Card" in their account settings. This is not an automatic memory, but a user-curated section. Key information like user name, professional role (e.g., "Mohamed, Software Testing Trainee"), recurring project names (e.g., "DeepSeek UI Project"), and personal preferences (e.g., "prefers Arabic language, likes detailed analysis") could be stored here.
This profile card could be automatically appended to the system prompt at the start of every new session, ensuring continuity without the user having to re-type anything. This idea has been discussed in community issues like #1189 and #1143.
Manual Context Reset/Truncation: Allow users to manually clear the current session's context with a simple command like /clear or via a UI button. This would give users the power to "flush" the conversation's memory without starting a brand new chat.
Pause & Resume: A feature that could be used in very long-running sessions to potentially compress and store the current context, allowing for a fresh start without losing history.
Advanced Long-Term Memory (Opt-In)
A more advanced, opt-in system where the model could, under the user's direction, automatically remember and recall specific facts across sessions (e.g., "Remember that I'm learning Software Testing to become a QA Lead."). This would transform DeepSeek from a stateless API into a true, personal long-term assistant, as requested by many users in issues like add Easydict application #67, #797, and #1106.
Enhanced File Management (Deleting/Managing Uploaded Files)
To help users manage the context window, there should be a clear interface to view, search, and delete images or files that have been uploaded during a session. Currently, once an image is uploaded, there is no direct way to remove its "weight" from the conversation from the user's side.
Community Validation (Supporting Evidence)
This request is not unique. It consolidates several popular feature requests from the DeepSeek community:
Issue #67: Implement Session Memory/Persistent Context.
Issue #75: Long-context handling & session memory (describes the token limit problem).
Issue #797: Persistent User Memory Across Sessions (includes a request to manage chats).
Issue #949: Context Memory Mode (suggests compressing conversation history).
Issue #1106 & #1189 & #1143: Various requests for long-term memory and user profile cards.
All these issues point to the same core need: users want DeepSeek to become a companion that learns and grows with them, rather than an isolated, stateless interaction.
Additional Context
As a power user who is actively testing the platform (as evidenced by my three previous feature requests #615, #616, #617), I believe these memory enhancements would be a game-changer. It would elevate the user experience from a simple "chatbot" to a true "persistent AI assistant."
I am excited about the upcoming V4 release and hope these features can be part of DeepSeek's evolution into a more personalized and intelligent platform.