Download Try Live

Memory Graph

The Memory Graph is an interactive network visualization that shows how your companion’s memories are semantically connected.

Accessing the Memory Graph

Open Settings > Memory > Graph, or click the brain icon on the main screen to go there directly. Choose Expand graph for more room. Escape or Collapse graph returns to the Memory page.

The same page has Facts for adding and deleting memories, Sessions for current turns and saved summaries, and Settings > Advanced for character state and parser testing.

Understanding the Visualization

Nodes (Memories)

Each node represents a stored memory (fact) about you, your relationship, or shared experiences.

Node Colors:

  • Blue — User facts (your preferences, background, attributes)
  • Pink — Relationship facts (dynamics between you and the companion)
  • Green — Shared experiences (events you’ve discussed together)

Connections

Lines between nodes indicate semantic similarity — memories that are related in meaning are connected. Animated particles flow along connections to visualize these relationships.

Statistics

The count below the graph shows the memories and connections in the current view.

Interactions

Selecting a Memory

Click a node or choose it from Inspect a memory:

  • The selected memory and its connections are highlighted
  • Unrelated memories fade
  • Details appear beside the graph on wide screens and below it on smaller screens
  • Open in Facts opens that exact saved memory in the inspector, where deletion requires confirmation

Filtering Categories

Use the category buttons above the graph to show or hide specific memory types. This helps focus on particular aspects of what your companion knows.

Reset View

Click “Reset view” to zoom out and see the full graph, clearing any selection.

Technical Details

The Memory Graph uses 384-dimensional embeddings (via Transformers.js with the multilingual paraphrase-multilingual-MiniLM-L12-v2 model) to compute semantic relationships between memories. Memories with a cosine similarity >= 0.5 are connected.

Reference Count tracks how many times a memory has been retrieved during conversations — higher counts indicate memories that frequently inform responses.

Importance Score (0-100) reflects how significant the memory is based on emotional content, personal details, and other heuristics.

Requirements

  • Memories must have embeddings to appear in the graph
  • The embedding model is loaded automatically on app startup
  • Existing memories without embeddings are backfilled automatically when the embedding model finishes loading