Skip to main content

Graph Visualizer

The Graph Visualizer is a high-performance, white-label administration hub for exploring and managing memories. It uses WebGL rendering (Sigma.js) to handle thousands of nodes and edges fluently.

Core Features​

  • Multi-Dimensional Navigation: Filter and explore the graph along platform-defined axes (e.g., "Tenants", "Cases", "Users").
  • Interactive Knowledge Graph: Zoom, pan, and drag nodes to explore semantic relationships.
  • Source-to-Memory Linking: A dual-pane view that shows the memory detail next to the original source document.
  • Semantic Highlighting: Precise coordinate-aware highlights on the source document (PDF) showing exactly where a memory was extracted from.

🔍 Side-by-Side Exploration​

When you select a memory in the graph, the Visualizer opens the Source Preview panel:

  1. Memory Detail: Shows the extracted text, structured data, and scores.
  2. Source PDF: Loads the original document.
  3. Coordinate Highlight: Automatically scrolls to and highlights the specific bounding boxes associated with the memory.

🎨 White-Labeling & Configuration​

The Visualizer is completely driven by a Platform Configuration JSON. Integrating platforms can customize:

  • Primary Dimensions: Define what users care about (Tenants, Clients, Patients).
  • Branding: Logo, brand colors, and theme.
  • Feature Toggles: Enable or disable Graph, Browser, Analytics, or Admin tools.
  • Custom Filters: Add domain-specific secondary filters (e.g., "Contract Value > $1M").

Example Configuration​

{
  "platformId": "procurement-pro",
  "name": "Procurement Memory Hub",
  "brandColor": "#7C3AED",
  "primaryDimensions": [
    {
      "id": "tenant",
      "label": "Tenant",
      "pluralLabel": "Tenants",
      "icon": "Briefcase",
      "mapping": {
        "type": "subject",
        "field": "tenant"
      }
    }
  ]
}

Advanced Tools​

Inside the Visualizer, admins can access specialized tools:

  • Quality Auditor: Find low-confidence or stale memories.
  • Duplicate Detector: Merge redundant pieces of knowledge.
  • Extraction Tester: Validate how the LLM maps conversations to Memory Blocks.