Skip to content

Commit b0aafaa

Browse files
sairin1202sairin1202
committed
update: multilang readme
1 parent b531d39 commit b0aafaa

7 files changed

Lines changed: 589 additions & 432 deletions

File tree

‎README.md‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -2,7 +2,7 @@
22

33
<div align="center">
44

5-
# MemU
5+
# memU
66

77
### Always-On Proactive Memory for AI Agents
88

‎readme/README_en.md‎

Lines changed: 96 additions & 71 deletions
Original file line numberDiff line numberDiff line change
@@ -2,7 +2,7 @@
22

33
<div align="center">
44

5-
# MemU
5+
# memU
66

77
### Always-On Proactive Memory for AI Agents
88

@@ -20,7 +20,9 @@
2020

2121
---
2222

23-
MemU is a **7×24 proactive memory framework** that continuously learns, anticipates, and adapts. It transforms passive LLM backends into intelligent agents with **always-on memory** that proactively surfaces insights, predicts needs, and evolves context without explicit queries.
23+
memU is a memory framework built for **24/7 proactive agents**.
24+
It is designed for long-running use and greatly **reduces the LLM token cost** of keeping agents always online, making always-on, evolving agents practical in production systems.
25+
memU **continuously captures and understands user intent**. Even without a command, the agent can tell what you are about to do and act on it by itself.
2426

2527
---
2628

@@ -32,105 +34,128 @@ If you find memU useful or interesting, a GitHub Star ⭐️ would be greatly ap
3234
---
3335

3436

35-
## ✨ Core Capabilities
37+
## ✨ Core Features
3638

3739
| Capability | Description |
3840
|------------|-------------|
39-
| 🔄 **Continuous Learning** | 7×24 memory extraction from every interaction—conversations, documents, actions |
40-
| 🎯 **Proactive Retrieval** | Anticipates information needs before being asked, surfaces relevant context automatically |
41-
| 🧠 **Context Evolution** | Memory structure adapts in real-time based on usage patterns and emerging topics |
42-
| 🔍 **Dual Intelligence** | Fast embedding-based recall + deep LLM reasoning for comprehensive understanding |
43-
| 🎨 **Multimodal Awareness** | Unified memory across text, images, audio, video—remembers what it sees and hears |
44-
41+
| 🤖 **24/7 Proactive Agent** | Always-on memory agent that works continuously in the background—never sleeps, never forgets |
42+
| 🎯 **User Intention Capture** | Understands and remembers user goals, preferences, and context across sessions automatically |
43+
| 💰 **Cost Efficient** | Reduces long-running token costs by caching insights and avoiding redundant LLM calls |
4544
---
4645

4746
## 🔄 How Proactive Memory Works
4847

49-
Unlike traditional retrieval systems that wait for queries, MemU operates in **continuous mode**:
48+
```bash
5049

51-
### Passive vs. Proactive Memory
50+
cd examples/proactive
51+
python proactive.py
5252

53-
| Traditional RAG | MemU Proactive Memory |
54-
|-----------------|----------------------|
55-
| ❌ Waits for explicit queries | ✅ Monitors context continuously |
56-
| ❌ Reactive information retrieval | ✅ Anticipates information needs |
57-
| ❌ Static knowledge base | ✅ Self-evolving memory structure |
58-
| ❌ One-time processing | ✅ Always-on learning pipeline |
53+
```
54+
55+
---
5956

6057
### Proactive Memory Lifecycle
6158
```
6259
┌─────────────────────────────────────────────────┐
63-
│ 1. CONTINUOUS INGESTION │
64-
│ └─ Every conversation, document, action │
65-
│ automatically processed 7×24 │
66-
└─────────────────────────────────────────────────┘
67-
↓
68-
┌─────────────────────────────────────────────────┐
69-
│ 2. REAL-TIME EXTRACTION │
70-
│ └─ Immediate memory item creation │
71-
│ No batch delays, instant availability │
72-
└─────────────────────────────────────────────────┘
73-
↓
74-
┌─────────────────────────────────────────────────┐
75-
│ 3. PROACTIVE STRUCTURING │
76-
│ └─ Auto-categorization into evolving topics │
77-
│ Hierarchical organization adapts to usage │
60+
│ 1. USER INITIAL QUERY │
61+
│ └─ User input, context, or any trigger event │
62+
│ Conversation starts here │
7863
└─────────────────────────────────────────────────┘
7964
↓
80-
┌─────────────────────────────────────────────────┐
81-
│ 4. ANTICIPATORY RETRIEVAL │
82-
│ └─ Surfaces relevant memory without prompting │
83-
│ Context-aware suggestions and insights │
84-
└─────────────────────────────────────────────────┘
65+
┌─────────────────────────────────────────────────┐
66+
│ 2. AGENT PLANNING / ACTIONS │
67+
│ └─ Analyze request, execute tasks │
68+
│ Retrieve relevant memories for context │
69+
└─────────────────────────────────────────────────┘
70+
↓
71+
┌─────────────────────────────────────────────────┐
72+
│ 3. MEMORIZE & UPDATE TODOLIST │
73+
│ └─ Store new insights, facts, preferences │
74+
│ Modify task list based on progress │
75+
└─────────────────────────────────────────────────┘
76+
↓
77+
┌─────────────────────────────────────────────────┐
78+
│ 4. PREDICT USER INTENT │
79+
│ └─ Anticipate next steps and needs │
80+
│ Proactively prepare relevant context │
81+
└─────────────────────────────────────────────────┘
82+
↓
83+
┌─────────────────────────────────────────────────┐
84+
│ 5. LOOP (2 → 4) │
85+
│ └─ Continuous iteration until task complete │
86+
│ Agent-driven proactive workflow │
87+
└─────────────────────────────────────────────────┘
8588
```
8689

8790
---
8891

8992
## 🎯 Proactive Use Cases
9093

91-
### 1. **Contextual Assistance**
92-
*Agent monitors conversation context and proactively surfaces relevant memories*
94+
### 1. **Information Recommendation**
95+
*Agent monitors interests and proactively surfaces relevant content*
9396
```python
94-
# User starts discussing a topic
95-
User: "I'm thinking about that project..."
96-
97-
# MemU automatically retrieves without explicit query:
98-
- Previous project discussions
99-
- Related preferences and constraints
100-
- Past decisions and their outcomes
101-
- Relevant documents and resources
102-
103-
Agent: "Based on your previous work on the dashboard redesign,
104-
I noticed you preferred Material UI components..."
97+
# User has been researching AI topics
98+
MemU tracks: reading history, saved articles, search queries
99+
100+
# When new content arrives:
101+
Agent: "I found 3 new papers on RAG optimization that align with
102+
your recent research on retrieval systems. One author
103+
(Dr. Chen) you've cited before published yesterday."
104+
105+
# Proactive behaviors:
106+
- Learns topic preferences from browsing patterns
107+
- Tracks author/source credibility preferences
108+
- Filters noise based on engagement history
109+
- Times recommendations for optimal attention
105110
```
106111

107-
### 2. **Predictive Preparation**
108-
*Agent anticipates upcoming needs based on patterns*
112+
### 2. **Email Management**
113+
*Agent learns communication patterns and handles routine correspondence*
109114
```python
110-
# Morning routine detection
111-
User logs in at 9 AM (usual time)
112-
113-
# MemU proactively surfaces:
114-
- Daily standup talking points
115-
- Overnight notifications summary
116-
- Priority tasks based on past behavior
117-
- Relevant context from yesterday's work
118-
119-
Agent: "Good morning! Here's what's relevant today..."
115+
# MemU observes email patterns over time:
116+
- Response templates for common scenarios
117+
- Priority contacts and urgent keywords
118+
- Scheduling preferences and availability
119+
- Writing style and tone variations
120+
121+
# Proactive email assistance:
122+
Agent: "You have 12 new emails. I've drafted responses for 3 routine
123+
requests and flagged 2 urgent items from your priority contacts.
124+
Should I also reschedule tomorrow's meeting based on the
125+
conflict John mentioned?"
126+
127+
# Autonomous actions:
128+
✓ Draft context-aware replies
129+
✓ Categorize and prioritize inbox
130+
✓ Detect scheduling conflicts
131+
✓ Summarize long threads with key decisions
120132
```
121133

122-
### 3. **Autonomous Memory Management**
123-
*System self-organizes without manual intervention*
134+
### 3. **Trading & Financial Monitoring**
135+
*Agent tracks market context and user investment behavior*
124136
```python
125-
# As interactions accumulate:
126-
✓ Automatically creates new categories for emerging topics
127-
✓ Consolidates related memories across modalities
128-
✓ Identifies patterns and extracts higher-level insights
129-
✓ Prunes outdated information while preserving history
130-
131-
# Result: Always-optimized memory structure
137+
# MemU learns trading preferences:
138+
- Risk tolerance from historical decisions
139+
- Preferred sectors and asset classes
140+
- Response patterns to market events
141+
- Portfolio rebalancing triggers
142+
143+
# Proactive alerts:
144+
Agent: "NVDA dropped 5% in after-hours trading. Based on your past
145+
behavior, you typically buy tech dips above 3%. Your current
146+
allocation allows for $2,000 additional exposure while
147+
maintaining your 70/30 equity-bond target."
148+
149+
# Continuous monitoring:
150+
- Track price alerts tied to user-defined thresholds
151+
- Correlate news events with portfolio impact
152+
- Learn from executed vs. ignored recommendations
153+
- Anticipate tax-loss harvesting opportunities
132154
```
133155

156+
157+
...
158+
134159
---
135160

136161
## 🗂️ Hierarchical Memory Architecture

0 commit comments

Comments
 (0)