About PlexusGraph
PlexusGraph is an experiment in understanding complex questions by building and analyzing knowledge graphs.
How it works
Each exploration starts with a question — something too tangled for a simple answer. An AI research agent then iteratively builds a weighted knowledge graph: nodes represent concepts (people, ideas, institutions, mechanisms) and edges represent relationships between them (enables, undermines, amplifies, triggers, etc.), each with a weight from 0-10.
The graph is built over multiple research iterations. Each round, the agent searches for real data, creates new concept nodes, discovers connections, and identifies gaps. After 10+ rounds, what emerges is a map of how a complex system actually works — not as a narrative, but as a structure.
Three layers of depth
Each exploration is presented at three levels:
- Summary — A plain-language explanation anyone can follow. What matters, and why.
- Interactive Graph — The full knowledge graph as a force-directed visualization. Drag, zoom, hover. See the structure for yourself.
- Deep Analysis — Feedback loops, central mechanisms, surprising connections, contradictions, and testable hypotheses extracted from the graph structure.
The tool
Explorations are built using BrainMCP, an open-source MCP server that provides a persistent weighted knowledge graph as long-term memory for LLMs. The raw graph data (SQLite databases) are available for download on each exploration page.
Why this exists
Most explanations of complex systems are either oversimplified narratives or impenetrable academic papers. Knowledge graphs sit in between — they preserve the complexity while making structure visible. When you can see that 30 mechanisms converge to block carbon pricing, or that a single feedback loop connects shadow banking to QE dependence, the picture changes.
The graphs aren't claims of truth. They're maps of how things connect, built from research, meant to be explored and questioned.