Skip to main content

​
Introduction to EngramDB

EngramDB is a specialized database system designed specifically for agent memory management. It provides efficient storage, retrieval, and querying of agent memories using a unified memory representation model.

​
What is an Engram?

In neuroscience, an engram is a hypothetical means by which memory traces are stored as biophysical or biochemical changes in the brain. EngramDB borrows this concept to represent agent memories as rich, multi-dimensional data structures that combine:
  • Vector embeddings for semantic content
  • Graph connections for relational information
  • Temporal layers for memory evolution
  • Flexible attributes for structured metadata

​
Why EngramDB?

Traditional databases are not optimized for the unique requirements of agent memory systems:
  • Vector databases excel at similarity search but lack graph capabilities
  • Graph databases handle relationships well but aren’t optimized for vector similarity
  • Document databases provide flexible schemas but lack specialized memory features
  • Time-series databases track changes over time but aren’t designed for semantic search
EngramDB combines the strengths of these different database types into a unified system specifically designed for agent memory management.

​
Key Use Cases

EngramDB is ideal for:
  • AI Agents: Store and retrieve memories for conversational agents, assistants, and autonomous systems
  • Knowledge Graphs: Build semantic knowledge graphs with vector similarity capabilities
  • Memory Evolution: Track how knowledge and beliefs change over time
  • Multi-modal Memory: Store and query memories across different modalities (text, images, etc.)
  • Cognitive Architectures: Implement memory systems for cognitive architectures

​
Core Features

  • Unified Memory Representation: Combines graph, vector, and temporal properties in a single primitive
  • Vector Similarity Search: Find memories with similar semantic content
  • Flexible Storage Options: In-memory database for testing and development, file-based for persistence
  • Query API: Rich querying with vector similarity, attribute filters, and temporal constraints
  • Memory Evolution: Track changes to memories over time with temporal layers
  • Python Bindings: First-class Python API for integration with ML and AI applications
  • Web Interface: Browser-based UI for visualization and interaction with the database

​
Architecture

EngramDB is built around these core components:
  1. MemoryNode: The fundamental unit of storage, combining vector embeddings, graph connections, temporal layers, and flexible attributes
  2. Storage Engines: Both in-memory and file-based storage options
  3. Vector Index: For efficient similarity search
  4. Query System: A fluent interface for building complex queries
  5. Database: A unified interface combining all components
  6. Language Bindings: Core implementation in Rust with Python bindings

​
Next Steps