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
MemU is an open-source memory framework for personalized AI companions. We treat memory as an agent-controlled file system that stores user information as documents. It provides high-speed memory retrieval and enables self-evolution without interacting with users. The framework is easily extendable, enabling you to create AI companions that genuinely remember and grow with their users.
MemU intelligently organizes memories into well-structured documents, creating a coherent knowledge base that grows with each interaction. Unlike traditional memory systems that store isolated fragments, MemU maintains contextual relationships and categorical organization.
79
88
89
+
**Key Benefits:**
90
+
- 🗂️ **Document-based storage** - Each memory type is stored as readable documents
91
+
- 🏷️ **Smart categorization** - Automatic classification by topics and contexts
92
+
- 🔍 **Easy retrieval** - Quick access to relevant information through semantic search
93
+
- 📊 **Structured format** - Consistent organization across all memory types
80
94
81
-
### 🔍 **Retrieval Fast**
82
-
95
+
---
83
96
97
+
## 🔗 **Linked Connections**
84
98
85
-
### 🎯 **Extension Framework**
86
-
99
+
<tablealign="center">
100
+
<tr>
101
+
<tdwidth="50%">
87
102
103
+
**Build a web of interconnected knowledge**
88
104
105
+
Different memories are interconnected through sophisticated semantic relationships, creating a rich network of contextual associations. This enables the AI to understand not just individual facts, but how concepts relate to each other across different conversations and contexts.
89
106
107
+
**Key Benefits:**
108
+
- 🕸️ **Semantic linking** - Memories connect based on meaning and context
109
+
- 🔄 **Cross-referencing** - Related information surfaces automatically
110
+
- 🧩 **Contextual understanding** - AI grasps relationships between different topics
111
+
- 📈 **Knowledge graph** - Dynamic network that expands with each interaction
**Self-improving AI that grows smarter over time**
96
136
137
+
The Memory Agent continuously performs Theory of Mind reasoning, analyzing stored memories to generate deeper insights about user preferences, behaviors, and needs. This creates an evolving understanding that enhances existing documents and generates new knowledge automatically.
138
+
139
+
**Key Benefits:**
140
+
- 🤔 **Theory of Mind** - Deep understanding of user mental states and preferences
- 💡 **Insight generation** - Creates new understanding from existing information
143
+
- 🌱 **Continuous growth** - Knowledge base becomes more sophisticated over time
144
+
145
+
</td>
146
+
</tr>
147
+
</table>
97
148
98
149
---
99
-
## 🤖 **Memory as documents**
100
-
150
+
151
+
## 🌫️ **Fade Memory**
152
+
153
+
<divalign="center">
154
+
155
+
**Natural forgetting that preserves what matters**
156
+
157
+
<table>
158
+
<tr>
159
+
<tdwidth="33%">
160
+
161
+
162
+
163
+
</td>
164
+
<tdwidth="34%">
165
+
166
+
167
+
168
+
</td>
169
+
<tdwidth="33%">
170
+
171
+
172
+
173
+
</td>
174
+
</tr>
175
+
</table>
176
+
177
+
</div>
178
+
179
+
MemU implements an intelligent LRU-like forgetting mechanism where memories naturally fade in importance without disappearing entirely. This mimics human memory patterns - rarely accessed information becomes less prominent while frequently referenced memories gain stronger relevance.
0 commit comments