​
Memory System Overview

Cipher’s memory system is a sophisticated, multi-layered architecture that enables AI agents to maintain persistent knowledge, learn from interactions, and continuously improve their reasoning capabilities. Unlike traditional stateless AI systems, cipher provides a comprehensive memory runtime that transforms single-pass interactions into intelligent, stateful conversations.

​
Storage Architecture

Cipher implements a four-layer storage architecture designed for different data types and access patterns:
  1. Cache Backend: Fast, ephemeral storage for temporary data and caching
  2. Database Backend: Persistent, reliable storage for long-term data
  3. Vector Storage: High-dimensional vector storage for semantic search
  4. Knowledge Graph: Graph database for entity relationships

​
Cache Backend

The cache backend provides fast, ephemeral storage for session data, temporary caches, and frequently accessed information. This layer prioritizes speed over persistence. Options:
  • Redis (Production): Distributed, high-performance cache with persistence options
  • In-Memory (Development): Simple local storage for testing and development
Use Cases:
  • Session state management
  • Temporary conversation context
  • Fast lookups and caching
  • Performance optimization
Configuration: 
# Cache Backend Configuration
STORAGE_CACHE_TYPE=redis              # redis, in-memory
STORAGE_CACHE_HOST=localhost          # Redis server hostname
STORAGE_CACHE_PORT=6379               # Redis server port
STORAGE_CACHE_PASSWORD=               # Redis authentication (optional)
STORAGE_CACHE_DATABASE=0              # Redis database number (0-15)

​
Database Backend

The database backend provides persistent, reliable storage for long-term data that must survive system restarts and maintain ACID properties. Options:
  • PostgreSQL (Production): Full-featured relational database with advanced features
  • SQLite (Development): Lightweight, file-based database for single-user applications
  • In-Memory (Testing): Temporary storage for testing scenarios
Use Cases:
  • Memory operation history and audit trails
  • User data and preferences
  • Configuration persistence
  • Long-term analytics and reporting
Configuration: 
# Database Backend Configuration
STORAGE_DATABASE_TYPE=sqlite          # sqlite, postgresql, in-memory
STORAGE_DATABASE_PATH=./data          # Directory path for SQLite
STORAGE_DATABASE_NAME=cipher.db       # Database filename

​
Memory Architecture

Built on top of these storage layers, Cipher implements a three-component memory system that provides intelligent knowledge management and continuous learning capabilities:

​
Knowledge Memory

Knowledge memory stores factual information, technical concepts, and domain-specific knowledge extracted from conversations. It uses vector embeddings to enable semantic search and retrieval. Content Types:
  • Technical concepts and programming patterns
  • Code snippets and implementation details
  • Domain knowledge and project context
  • Best practices and optimization techniques
Overview: Storage: Vector embeddings in dedicated collections (Qdrant, Milvus, or in-memory)

​
Reflection Memory

Reflection memory captures the agent’s reasoning processes and decision-making patterns, enabling continuous learning and improvement through pattern recognition. Content Types:
  • Step-by-step reasoning traces
  • Decision-making processes and logic
  • Pattern recognition and effectiveness metrics
  • Learning experiences and quality evaluations
Overview: Storage: Separate vector collection optimized for reasoning pattern analysis

​
Knowledge Graph

Knowledge graph provides advanced relationship modeling between entities, enabling complex queries and deep understanding of connections within the knowledge base. Content Types:
  • Entity relationships and connections
  • Hierarchical knowledge structures
  • Cross-domain knowledge mapping
  • Complex query patterns
Overview: Storage: Graph database (Neo4j or in-memory) for relationship modeling

​
Vector Storage Configuration

Both knowledge and reflection memory utilize vector storage for semantic search capabilities: 
# Vector Store Configuration
VECTOR_STORE_TYPE=qdrant                   # qdrant, milvus, in-memory
VECTOR_STORE_HOST=localhost                # Vector database hostname
VECTOR_STORE_PORT=6333                     # Vector database port
VECTOR_STORE_COLLECTION=knowledge_memory   # Knowledge collection name
VECTOR_STORE_DIMENSION=1536                # Embedding dimensions

# Reflection Memory Configuration
REFLECTION_VECTOR_STORE_COLLECTION=reflection_memory  # Reflection collection
DISABLE_REFLECTION_MEMORY=false            # Enable/disable reflection memory

# Memory Search Configuration
SEARCH_MEMORY_TYPE=both                    # knowledge, reflection, both

​
Knowledge Graph Configuration

Knowledge graph functionality provides advanced relationship modeling: 
# Knowledge Graph Configuration
KNOWLEDGE_GRAPH_ENABLED=true              # Enable knowledge graph features
KNOWLEDGE_GRAPH_TYPE=neo4j                # neo4j, in-memory
KNOWLEDGE_GRAPH_HOST=localhost            # Neo4j hostname
KNOWLEDGE_GRAPH_PORT=7687                 # Neo4j bolt port
KNOWLEDGE_GRAPH_USERNAME=neo4j            # Neo4j username
KNOWLEDGE_GRAPH_PASSWORD=                 # Neo4j password (required)

​
Memory Tools

Cipher provides internal tools for memory operations across all three memory components:

​
Knowledge Memory Tools

  • cipher_extract_and_operate_memory: Extract and manage knowledge entries
  • cipher_memory_search: Search stored knowledge using semantic queries

​
Reflection Memory Tools

  • cipher_store_reasoning_memory: Store reasoning patterns and evaluations
  • cipher_extract_reasoning_steps: Extract reasoning traces from conversations
  • cipher_evaluate_reasoning: Evaluate reasoning quality and effectiveness
  • cipher_search_reasoning_patterns: Search for similar reasoning patterns

​
Knowledge Graph Tools

  • cipher_add_node: Add nodes to the knowledge graph with labels and properties
  • cipher_add_edge: Create relationships between nodes in the knowledge graph
  • cipher_search_graph: Search for nodes and edges with filtering capabilities
  • cipher_get_neighbors: Get neighboring nodes and their relationships
  • cipher_extract_entities: Extract entities from text and add to knowledge graph
  • cipher_update_node: Update existing node properties and labels
  • cipher_delete_node: Remove nodes and their relationships from the graph
  • cipher_query_graph: Execute custom queries against the knowledge graph
  • cipher_enhanced_search: Advanced search with semantic and fuzzy matching
  • cipher_intelligent_processor: Process natural language to manage entities automatically
  • cipher_relationship_manager: Intelligently manage complex relationship operations

​
Memory System Benefits

​
Persistent Intelligence

  • Cross-Session Knowledge: Information persists across conversations and sessions
  • Contextual Understanding: Build comprehensive domain and project knowledge
  • Continuous Learning: Improve through pattern recognition and quality evaluation

​
Advanced Capabilities

  • Semantic Search: Natural language queries across stored knowledge
  • Pattern Recognition: Identify and reuse successful reasoning approaches
  • Relationship Modeling: Understand complex connections between concepts
  • Quality Assurance: Automated evaluation and improvement of memory content

​
Next Steps

Explore the detailed capabilities of each memory component: The memory system transforms cipher from a simple conversation tool into a sophisticated AI agent with persistent intelligence, continuous learning capabilities, and comprehensive knowledge management.