Knowledge Graph

The Knowledge Graph is an auto-categorizing knowledge base that stores, links, and visualizes information with semantic connections. It serves as the long-term memory for all agents.

How It Works

When agents generate research, reports, or artifacts, the Knowledge Graph agent automatically:

1. Ingests the raw content into .magent/vault/raw/
2. Extracts key entities, topics, and relationships
3. Creates semantic links between related knowledge entries
4. Categorizes entries by domain, type, and confidence level
5. Makes everything searchable and retrievable by other agents

Vault Structure

.magent/vault/
  raw/          Original ingested documents
  wiki/         Processed knowledge articles
  outputs/      Generated artifacts and deliverables

Entry Format

Each knowledge entry is stored as a JSON file with metadata:

{
  "id": "kg-a1b2c3d4",
  "title": "Competitive Analysis: SaaS Market Q1 2026",
  "category": "research",
  "tags": ["saas", "market-analysis", "competitive"],
  "created": "2026-05-20T14:30:00Z",
  "source_agent": "researcher",
  "connections": ["kg-e5f6g7h8", "kg-i9j0k1l2"],
  "confidence": 0.92,
  "content": "..."
}

Radial Visualization

The dashboard provides an interactive radial graph that displays knowledge entries as nodes with connections drawn between semantically related items. Features include:

• Color coding — Nodes colored by category (research, technical, business, creative)
• Size weighting — Larger nodes indicate higher connection count or confidence
• Interactive exploration — Click nodes to expand connections, hover for previews
• Search and filter — Find entries by keyword, category, or date range

Agent Integration

The Knowledge Graph offers read and write access to any agent:

• vault_read — Search and retrieve knowledge entries by query or ID
• vault_write — Store new entries with automatic categorization
• vault_link — Create explicit connections between entries

The Orchestrator automatically queries the Knowledge Graph for relevant context before delegating tasks, giving agents access to previously gathered intelligence.

Dashboard View

The Knowledge Graph view in the dashboard provides:

• Radial graph visualization with pan/zoom controls
• List view with sortable columns (title, category, date, connections)
• Full-text search across all entries
• Category filters and date range selectors
• Entry detail panel with content preview and connection map
• Manual entry creation and editing

Frequently Asked Questions

What is the Knowledge Graph?

An auto-categorizing knowledge base that stores, links, and visualizes information with semantic connections. It serves as the long-term memory for all agents.

How does content get into the Knowledge Graph?

The Knowledge Graph agent ingests raw content into .magent/vault/raw/, extracts key entities, topics, and relationships, creates semantic links between related entries, categorizes them by domain, type, and confidence level, and makes everything searchable and retrievable by other agents.

What is stored in a knowledge entry?

Each entry is a JSON file with an id, title, category, tags, creation date, source agent, connections to related entries, a confidence score, and the content itself.

Can any agent access the Knowledge Graph?

Yes. The Knowledge Graph offers read and write access to any agent through vault_read, vault_write, and vault_link. The Orchestrator also automatically queries it for relevant context before delegating tasks.

What does the radial visualization show?

An interactive graph of knowledge entries as nodes, color-coded by category, sized by connection count or confidence, with connections drawn between semantically related items. Click a node to expand its connections or hover for a preview.

Can I search and filter knowledge entries?

Yes. The dashboard provides full-text search across all entries, category filters, date range selectors, and a sortable list view by title, category, date, and connections.