NeuroGraph Cognitive Memory

Metadata

name: neurograph
version: 0.4.0
autoload: true

Description

NeuroGraph provides dynamic cognitive memory for the OpenClaw AI assistant.
It combines three memory layers:

• Semantic Memory (Vector DB) — content similarity via embeddings
• Structural Memory (Hypergraph) — N-ary concept relationships
• Temporal Memory (Spiking Neural Network) — causal learning via STDP

Environment Variables

• NEUROGRAPH_WORKSPACE_DIR — Working directory for checkpoints and data (default: ~/NeuroGraph/data)

Capabilities

• Automatic content ingestion (text, code, markdown, URLs, PDFs)
• STDP-based causal learning across conversations
• Semantic similarity recall
• Prediction generation and surprise-driven exploration
• Cross-session persistence via checkpoints

Integration

NeuroGraph auto-loads as an OpenClaw skill. On each message:

1. Content is ingested through the 5-stage pipeline
2. The SNN runs one learning step (STDP + homeostatic + structural plasticity)
3. Predictions are evaluated and surprise exploration triggered
4. State auto-saves every 10 messages

CLI

Use feed-syl for manual ingestion and status:

feed-syl --status              # Show graph statistics
feed-syl --text "content"      # Ingest text
feed-syl --file path/to/file   # Ingest a file
feed-syl --workspace           # Ingest OpenClaw workspace docs
feed-syl --query "search term" # Semantic search

Files

• neuro_foundation.py — Core SNN engine with STDP, hypergraph, predictions
• universal_ingestor.py — 5-stage ingestion pipeline
• openclaw_hook.py — OpenClaw integration singleton