3D Memory Field
Semantic navigation through the Nosos vault
What Is This?
The 3D Memory Field is a spatial interface to the Nosos vault. Every note, transcript, and session exists as a point in 3D space — positioned not by folder hierarchy, but by semantic meaning.
“Navigate memories in three dimensions.”
How It Works
1. Vector Embeddings
Every file in the vault is converted to a high-dimensional vector using sentence-transformers (all-MiniLM-L6-v2). Similar content = similar vectors.
2. Dimensionality Reduction
PCA reduces embeddings to 3D coordinates (X, Y, Z). Files that are semantically related cluster together in space.
3. LanceDB Storage
Vectors and coordinates are stored in ~/.openclaw/workspace/memory_vectors_3d/ for fast semantic search.
The Index
440+ chunks indexed
├── 🧬 Core/ Identity, soul, system (priority 9-10)
├── 🧠 Cognition/ Fallacies, frameworks, models (priority 7)
├── 🔧 Projects/ Active work, KBIRD, tools (priority 6)
├── 🕸️ Graph/ Navigation MOCs (priority 5)
├── 📅 Time/ Daily logs, sessions (priority 4)
└── 🎙️ Transcripts/ Podcasts, recordings (priority 4)
Search command:
python3 ~/.openclaw/workspace/scripts/memory-search.py "your query" 53D Visualization
The 3D coordinate data lives in:
- JSON:
~/.openclaw/workspace/memory_3d.json - Database:
~/.openclaw/workspace/memory_vectors_3d/ - Prototype:
~/.openclaw/workspace/memory_3d_prototype.py
To Regenerate
cd ~/.openclaw/workspace
python3 scripts/memory_3d_prototype.pySpatial Relationships
In the 3D field:
- Distance = semantic dissimilarity
- Clusters = topic neighborhoods
- Time = one axis (usually)
- Depth = abstraction level
Future: Interactive 3D View
Planned: A web-based 3D scatter plot where you can:
- Fly through the memory space
- Click points to open files
- Search and highlight related content
- See temporal drift over time
Related
- Memory System — How we persist
- Juggling Framework — The philosophy of dynamic memory
- Tools — Infrastructure details
The vault is not a hierarchy. It’s a field. 🌐