Knowledge Graph Core

The knowledge graph is a semantic layer over wiki content that enables AI agents to discover relationships, traverse connections, and propose new knowledge — all grounded in human-authored content.

Architecture

Four PostgreSQL tables form the graph:

Sub-Features

• GraphProjector — PageFilter that synchronizes frontmatter to the graph on every page save
• Knowledge Proposals — Proposal/approval/rejection workflow with frontmatter write-back
• Knowledge Admin UI — Three-tab admin panel for proposals, node explorer, and edge explorer
• Provenance Model — Three-tier trust model (human-authored → ai-inferred → ai-reviewed)
• Frontmatter Conventions — Rules for how frontmatter keys become properties vs. edges

MCP Integration

Two separate MCP endpoints serve different use cases:

• /wikantik-admin-mcp (authoring + curation, 27 tools on the server overall). The KG-facing subset is propose_knowledge, list_proposals, inspect_proposals, review_proposals, curate_nodes, curate_edges, list_orphaned_kg_nodes, plus admin-bypass copies of query_nodes and search_knowledge so curators see entities they have just created.
• /knowledge-mcp (consumption, 21 read-only tools on the server overall). The KG-facing subset is discover_schema, query_nodes, traverse, get_node, search_knowledge, and find_similar.

The authoring endpoint is /wikantik-admin-mcp. An older /mcp path appears in early notes; it no longer exists. Curation writes are refused when an edge violates a SHACL shape.