MotteMB provides sophisticated memory management using vector embeddings for context-aware agent memory and semantic retrieval with OpenAI's text-embedding-3-large model.
1536-dimensional vectors using OpenAI's text-embedding-3-large model for semantic understanding.
Find relevant memories using natural language queries with cosine similarity matching.
PostgreSQL with pgvector extension for efficient vector storage and indexing with HNSW algorithms.
Import/export memories in JSON, CSV, TXT, and JSONL formats with batch processing capabilities.
{
"id": "mem_abc123",
"content": "Customer refund policy allows returns within 30 days...",
"embedding": [0.123, -0.456, 0.789, ...], // 1536 dimensions
"metadata": {
"category": "policy",
"source": "handbook",
"priority": "high",
"tags": ["refund", "policy", "customer-service"],
"created_at": "2024-01-15T10:30:00Z",
"updated_at": "2024-01-15T10:30:00Z"
},
"similarity_score": 0.87 // Only present in search results
}| Algorithm | Use Case | Performance | Accuracy |
|---|---|---|---|
| Cosine Similarity | General semantic search | Fast | High |
| HNSW Index | Large-scale retrieval | Very Fast | High |
| Exact Search | Small datasets | Slow | Perfect |
Automatically identify and remove duplicate or near-duplicate memories using similarity thresholds >0.95.
Rebuild HNSW indexes for optimal query performance with configurable ef_construction and M parameters.
Compress vector storage using quantization techniques to reduce memory usage by up to 75%.
Process large datasets efficiently with chunked embedding generation and parallel database operations.
POST /api/memory/store
Content-Type: application/json
{
"content": "Customer refund policy allows returns within 30 days for digital products and 60 days for physical products.",
"metadata": {
"category": "policy",
"source": "handbook",
"priority": "high",
"tags": ["refund", "policy", "customer-service"]
}
}Stores a new memory with automatic embedding generation and metadata indexing.
POST /api/memory/search
Content-Type: application/json
{
"query": "What is our refund policy for digital products?",
"limit": 5,
"threshold": 0.7,
"filters": {
"category": "policy",
"priority": ["high", "medium"]
}
}Searches for relevant memories using semantic similarity with optional metadata filters.
POST /api/memory/import Content-Type: multipart/form-data file: [uploaded file] format: "json" | "csv" | "txt" | "jsonl" batch_size: 100 (optional) auto_categorize: true (optional)
Bulk import memories from various file formats with automatic processing and categorization.
GET /api/memory/stats
Response:
{
"total_memories": 1250,
"categories": {
"policy": 45,
"faq": 120,
"procedures": 85
},
"storage_size": "2.3 MB",
"index_size": "1.1 MB",
"last_optimization": "2024-01-15T10:30:00Z"
}Get comprehensive statistics about your memory bank including storage usage and categorization.