Upload README.md with huggingface_hub

README.md

@@ -2,118 +2,87 @@
 title: Mnemo v2 - AI Memory System
 emoji: 🧠
 colorFrom: purple
-colorTo: indigo
+colorTo: blue
 sdk: gradio
 sdk_version: 5.9.1
 app_file: app.py
 pinned: false
 license: mit
-short_description: "AI memory with real embeddings & temporal decay"
+short_description: AI memory with semantic embeddings & temporal decay
 tags:
   - memory
   - ai-memory
   - llm-memory
-  - memory-system
   - semantic-memory
-  - long-term-memory
-  - conversational-memory
-  - chatbot-memory
-  - agent-memory
-  - rag
-  - retrieval
-  - semantic-search
-  - vector-search
-  - embeddings
   - sentence-transformers
   - faiss
   - hnsw
   - temporal-decay
-  - langchain
-  - llamaindex
   - mem0
-  - memgpt
-  - zep
   - mcp
-  - tool
+  - rag
+  - retrieval
+  - embeddings
+  - vector-search
+  - zerogpu
 ---
 
-# 🧠 Mnemo v2 - Enhanced Memory for LLMs
+# 🧠 Mnemo v2 - Enhanced Memory System for LLMs
 
-**Production-ready memory system with real embeddings, HNSW search, and temporal decay.**
+Production-ready memory layer for AI applications with **real semantic embeddings**, **O(log n) vector search**, and **temporal decay**.
 
-## ✨ Key Features
+## ✨ Features
 
-| Feature | Implementation |
-|---------|----------------|
-| **Real Embeddings** | sentence-transformers `all-MiniLM-L6-v2` (384 dim) |
-| **Fast Search** | FAISS HNSW index - O(log n) complexity |
+| Feature | Description |
+|---------|-------------|
+| **Semantic Embeddings** | `sentence-transformers/all-MiniLM-L6-v2` (384 dimensions) |
+| **Vector Search** | FAISS HNSW index with O(log n) complexity |
 | **Temporal Decay** | Ebbinghaus forgetting curve (configurable half-life) |
-| **Smart Scoring** | 40% similarity + 30% recency + 30% frequency |
-| **Auto Dedup** | ADD/UPDATE detection (0.85 similarity threshold) |
-| **Multi-User** | User ID scoping for multi-tenant apps |
+| **Relevance Scoring** | Composite: 40% similarity + 30% recency + 30% frequency |
+| **Smart Deduplication** | Automatic UPDATE detection for similar memories |
+| **Multi-User** | Full user_id scoping with per-user indices |
+| **ZeroGPU** | Dynamic H200 GPU allocation for fast inference |
 
 ## 🚀 Quick Start
 
-```python
-from mnemo_core import MnemoV2
+1. **Load Examples**: Click "Load Examples" to populate test memories
+2. **Search**: Ask semantic queries like "What are the user's preferences?"
+3. **Add**: Store new memories with importance weights and tags
+4. **Manage**: List, delete, or clear memories
 
-memory = MnemoV2(decay_half_life_days=7.0)
-
-# Add memories
-memory.add("User prefers Python for backend development", importance=0.9)
-memory.add("Meeting with team every Monday at 10am")
-
-# Search with relevance scoring
-results = memory.search("What programming language does the user prefer?")
-for r in results:
-    print(f"{r['content']} (relevance: {r['relevance_score']:.2f})")
-```
-
-## 📊 Architecture
+## 🔬 Architecture
 
 ```
-Input Text ──► Embedding ──► HNSW Index ──► Candidates
-                (384 dim)     (FAISS)           │
-                                                ▼
-                                      Relevance Scoring
-                                      ─────────────────
-                                      • Similarity (40%)
-                                      • Recency (30%) ◄── Temporal Decay
-                                      • Frequency (30%)
-                                      • Importance mult
-                                                │
-                                                ▼
-                                          Ranked Results
-```
-
-## 🔧 Configuration
-
-```python
-memory = MnemoV2(
-    decay_half_life_days=7.0,      # Memory decay rate
-    similarity_weight=0.4,          # Weight for semantic similarity
-    recency_weight=0.3,             # Weight for recency
-    frequency_weight=0.3,           # Weight for access frequency
-    update_threshold=0.85,          # Similarity for UPDATE vs ADD
-    max_memories_per_user=10000     # Auto-pruning limit
-)
+Query → Sentence-Transformer → 384-dim Embedding
+                                    │
+                                    ▼
+                            FAISS HNSW Index
+                            (M=32, efSearch=50)
+                                    │
+                                    ▼
+                           Relevance Scoring
+                           ├─ 40% Similarity
+                           ├─ 30% Recency (decay)
+                           └─ 30% Frequency
+                                    │
+                                    ▼
+                            Ranked Results
 ```
 
-## 📈 v1 vs v2 Comparison
+## 📊 Performance
 
-| Metric | v1 (Hash) | v2 (Transformer) |
-|--------|-----------|------------------|
-| Embedding Quality | Poor (hash-based) | Excellent (384-dim semantic) |
-| Search Complexity | O(n) linear | O(log n) HNSW |
-| Temporal Awareness | None | Ebbinghaus decay |
-| Relevance Scoring | Similarity only | Composite (3 factors) |
-| Deduplication | None | Auto UPDATE detection |
+| Metric | Value |
+|--------|-------|
+| Search Latency (GPU) | ~15-30ms |
+| Embedding Dimensions | 384 |
+| Index Type | HNSW Flat |
+| Decay Half-Life | 7 days (configurable) |
 
-## 🔗 Links
+## 🔗 Related
 
-- **API Server:** [AthelaPerk/mnemo-mcp](https://huggingface.co/spaces/AthelaPerk/mnemo-mcp)
-- **Source Code:** [AthelaPerk/mnemo-memory](https://huggingface.co/AthelaPerk/mnemo-memory)
+- **MCP Server**: [AthelaPerk/mnemo-mcp](https://huggingface.co/spaces/AthelaPerk/mnemo-mcp)
+- **Inspired by**: [mem0](https://mem0.ai)
 
-## License
+## 📜 License
 
-MIT License
+MIT License - free for commercial use.