Skip to main content

​
Overview

MemU memories are more than just text blobs. Each memory carries identity fields — structured metadata that captures the emotional, relational, and contextual dimensions of information. This enables the agent to recall not just what happened, but how it felt, who was involved, and why it mattered.

​
Entity Fields

​
Entities

A JSON array of people, places, organizations, and things referenced in the memory: 
{
  "entities": ["Jason", "Richard", "Dell R750", "Titanium Computing"]
}
Entities are used in:
  • Search boosting: Queries mentioning an entity boost memories about that entity
  • Relationship mapping: Understanding who is connected to whom
  • Context assembly: When the agent is talking to a specific user, relevant entity memories are prioritized

​
Relationships

Memories can encode relationships between entities: 
{
  "content": "Jason and Richard are business partners at Titanium Computing",
  "entities": ["Jason", "Richard", "Titanium Computing"],
  "type": "relationship"
}

​
Emotional Context

The emotional_context field captures the emotional tone or significance of an interaction: 
{
  "content": "Jason was frustrated with the WhatsApp connection dropping repeatedly",
  "emotional_context": "frustrated, technical difficulty",
  "significance": 6
}
This helps the agent:
  • Recall how a user felt about something
  • Avoid repeating approaches that caused frustration
  • Show appropriate empathy in responses

​
Memory Types

The type field categorizes memories:
TypeDescriptionExample
factObjective information”Jason has 30+ years IT experience”
preferenceUser preferences”Prefers open-source when quality is comparable”
eventSomething that happened”Deployed the new dashboard on Feb 5”
relationshipPeople connections”Jason and Richard are business partners”
instructionHow-to knowledge”To rebuild native modules: pnpm rebuild better-sqlite3”
observationAgent’s own observations”User tends to work late on Thursdays”

​
Significance Interaction

Identity fields interact with significance scoring. A memory with:
  • High emotional intensity -> higher significance
  • Multiple entities -> likely more important
  • Relationship type -> higher baseline significance
The agent learns to calibrate significance based on these signals over time.

​
Querying by Identity

Memory recall can leverage identity fields for more targeted searches: 
{
  "tool": "memory_recall",
  "input": {
    "query": "Richard preferences",
    "entity": "Richard",
    "type": "preference"
  }
}

​
Privacy Considerations

Identity fields can contain sensitive personal information. All memory data is stored locally in ~/.argentos/memory.db and never transmitted externally. The agent’s memory is private to the installation.