Skip to main content

​
Memory System Infrastructure

The memory system provides both short-term and long-term memory capabilities for all MOOD MNKY agents, enabling context retention and personalized experiences.

​
Overview

The memory system enables agents to:
  • Remember conversations - Maintain context across sessions
  • Learn preferences - Build user profiles over time
  • Retain knowledge - Store important information
  • Personalize interactions - Use historical data for customization

​
Architecture

​
Short-Term Memory

Short-term memory handles immediate context:
  • Conversation Buffer: Current conversation context
  • Working Memory: Active task information
  • Session State: Temporary session data
  • Cache: Frequently accessed information

​
Long-Term Memory

Long-term memory stores persistent information:
  • User Profiles: Comprehensive user data and preferences
  • Conversation History: Archived conversations
  • Knowledge Base: Learned facts and information
  • Preferences: User preferences and settings

​
Memory Types

​
Episodic Memory

Stores specific events and experiences:
  • Conversation transcripts
  • User interactions
  • Event sequences
  • Temporal information

​
Semantic Memory

Stores general knowledge:
  • Facts and concepts
  • Relationships
  • Patterns and rules
  • Domain knowledge

​
Procedural Memory

Stores how-to information:
  • Task procedures
  • Workflow patterns
  • Best practices
  • Operational knowledge

​
Memory Processes

Memory processing flow diagram showing encoding, storage, and retrieval processes

​
Encoding

Information is processed and encoded:
  • Extraction: Key information extraction
  • Normalization: Data standardization
  • Embedding: Vector representation
  • Indexing: Search index creation

​
Storage

Encoded information is stored:
  • Vector Database: Semantic embeddings
  • Relational Database: Structured data
  • File Storage: Documents and media
  • Cache: Fast access layer

​
Retrieval

Information is retrieved when needed:
  • Semantic Search: Similarity-based retrieval
  • Query Processing: Structured queries
  • Context Filtering: Relevance filtering
  • Ranking: Result prioritization

​
Configuration

Memory system configuration: 
memory:
  short_term:
    buffer_size: 4096
    ttl: 3600  # 1 hour
  long_term:
    vector_store: "pinecone"
    database: "postgresql"
    retention_days: 365