Knowledge Visualization
PiSovereign includes an interactive 3D knowledge graph that lets you explore how your AI assistant’s memory works — watching knowledge form, connect, and naturally decay over time.
Accessing the Knowledge Graph
Navigate to Settings → Knowledge Graph (or visit /knowledge directly). The visualization loads on demand — the Three.js library (~600 KB) is only fetched when you open this page.
What You’re Looking At
The knowledge graph renders your AI’s memory as an interactive 3D network of nodes (memories) and edges (connections between them).
Node Colors
| Color | Memory Type | What It Stores |
|---|---|---|
| 🔵 Blue | Episodic | Specific conversations and events — “what happened” |
| 🟢 Green | Semantic | Facts and knowledge — “what is true” |
| 🟠 Orange | Procedural | Skills and procedures — “how to do things” |
Node Size
Larger nodes are more important — they’ve been accessed more frequently or connected to more other memories.
Node Opacity
Opacity shows retention — how well the memory is preserved:
- Bright / fully opaque: Recently accessed, well-retained
- Fading / translucent: Decaying — hasn’t been accessed in a while
- Hidden: Below 10% retention — automatically hidden to reduce clutter
This uses the Ebbinghaus forgetting curve: memories fade exponentially but are reinforced each time they’re accessed (spaced repetition).
Edges
Lines connecting nodes represent causal or semantic relationships:
- Thickness: Stronger causal relationship = thicker line
- Animation: Animated edges show the direction of causal influence
Interacting with the Graph
Hover
Move your cursor over any node to see a tooltip with:
- Name: Short description of the memory
- Type: Episodic, Semantic, or Procedural
- Retention: Current retention percentage (e.g., “78%”)
- Source: Where the memory came from (conversation, email, calendar, etc.)
- Created: When the memory was first stored
Rotate and Zoom
- Click and drag: Rotate the 3D view
- Scroll wheel: Zoom in and out
- Right-click drag: Pan the view
Real-Time Updates
The graph updates in real-time via server-sent events (SSE). You’ll see:
- New nodes appearing when memories are created during conversations
- Nodes fading as time-based decay progresses
- New edges forming as the AI discovers connections between memories
Performance
The graph is designed to run smoothly even on lower-powered devices:
- Maximum 5,000 nodes displayed at once (most important nodes prioritized)
- Nodes below 10% retention are hidden automatically
- The Three.js renderer uses WebGL for hardware-accelerated graphics
- On devices without WebGL support, a simplified 2D fallback is used
Configuration
Administrators can adjust these settings in config.toml:
[knowledge_visualization]
enabled = true # Enable/disable the visualization page
max_nodes = 5000 # Maximum nodes rendered
min_retention_threshold = 0.1 # Hide nodes below this retention (0.0–1.0)
sse_enabled = true # Real-time updates via SSE
See Also
- Features Overview — High-level feature summary
- Innovation Features — Technical deep-dive (#19: Temporal Knowledge Decay Visualization)
- Memory System — How the RAG pipeline and memory stores work