Dream Deep Memory
An open-source AI memory plugin that uses structured memory and an active injection mechanism to achieve long-term memory management across sessions. It supports seven categories of memory classification and offline dream-based consolidation.
Timeline (last 90 days)
The open-source memory plugin "Dream Deep Memory" has version number v0.38.3.
"Dream Deep Memory" is an open-source plugin released on GitHub by a Bilibili creator.
"Dream Deep Memory" divides memories into seven categories: skills, soul, users, projects, topics, facts, lessons and guidelines.
"Dream Deep Memory" adopts an approach of structured memory plus proactive injection: it first distills conversations into entries, then presents them on demand when a window opens.
The "dream" function of "Dream Deep Memory" is an offline batch-processing-style memory consolidation in which another model instance summarizes, deduplicates, merges and annotates historical conversations during idle periods.
The "dream" integration process of "Dream Deep Memory" takes about two hours.
The offline memory integration of "Dream Deep Memory" requires additional model call costs and depends on a scheduling strategy for deciding when to trigger integration.
User-type memories (preferences, taboos) should be loaded in almost every conversation, while project-type memories are isolated by workspace.
Lesson and guideline memories are high-priority behavioral constraints and often need to be injected into the prompt before generation.
This user accumulated 1,612 memories and 113 dreams over five weeks, of which 11 have been marked as "forgotten" but not deleted.
3 more timeline events
All Facts (13)
"Dream Deep Memory" provides a visual "mind map" index, with "me" at the center and more than a hundred projects around it; each node corresponds to a memory, and its retention time and conversation context can be viewed.
50%Unverified"Dream Deep Memory" is an open-source plugin released on GitHub by a Bilibili creator.
50%Unverified"Dream Deep Memory" divides memories into seven categories: skills, soul, users, projects, topics, facts, lessons and guidelines.
50%Unverified"Dream Deep Memory" adopts an approach of structured memory plus proactive injection: it first distills conversations into entries, then presents them on demand when a window opens.
50%UnverifiedThe "dream" function of "Dream Deep Memory" is an offline batch-processing-style memory consolidation in which another model instance summarizes, deduplicates, merges and annotates historical conversations during idle periods.
50%UnverifiedThe "dream" integration process of "Dream Deep Memory" takes about two hours.
50%UnverifiedThe offline memory integration of "Dream Deep Memory" requires additional model call costs and depends on a scheduling strategy for deciding when to trigger integration.
50%UnverifiedUser-type memories (preferences, taboos) should be loaded in almost every conversation, while project-type memories are isolated by workspace.
50%UnverifiedLesson and guideline memories are high-priority behavioral constraints and often need to be injected into the prompt before generation.
50%UnverifiedThis user accumulated 1,612 memories and 113 dreams over five weeks, of which 11 have been marked as "forgotten" but not deleted.
50%Unverified"Dream Deep Memory" uses "forgetting marks" instead of deletion, so memories fade but are not removed.
50%UnverifiedIf you give "Dream Deep Memory"'s GitHub address to DSH, DSH can complete the installation automatically.
50%UnverifiedThe open-source memory plugin "Dream Deep Memory" has version number v0.38.3.
50%