KC-Wiki Agentic Knowledge Platform
Agent-accessible wiki with governed ingest, review, and MCP integration
Built KC-Wiki as the knowledge and memory layer of the KC Enterprise AI platform: a custom MediaWiki instance with LDAP authentication, automated document-to-wiki pipelines routed through LiteLLM guardrails, human-in-the-loop review for sensitive updates, and a MediaWiki MCP server so Lux GPT and other agents can search, cite, and publish grounded content on the shared platform network.
Client
Luxottica
Completion
Ongoing
Category
Secure AI & Automation
Also: Identity & Access
Part of platform
KC Enterprise AI & Operations Platform →Situation
Retail IT knowledge lived in scattered files with no agent-safe access or publish controls. The platform needed durable, citable documentation that AI workflows could read and update under governance—not another static wiki.
Task
Deliver an agent-native knowledge platform that ingests documents, applies guardrailed synthesis via LiteLLM, publishes to MediaWiki with reviewer-approved workflows, and exposes wiki content through MediaWiki MCP on the shared production network.
Action
→Deployed LDAP-authenticated MediaWiki as the system of record, integrated with the shared LiteLLM proxy
→Built automated document ingest with guardrailed LiteLLM synthesis, semantic diff, auto-publish, and reviewer queue
→Implemented doc-extract HTTP sidecar for DOCX/PDF processing where hardened n8n images lack Python
→Shipped MediaWiki MCP for agent and chat tool access—grounded Q&A and governed write-back from Lux GPT
→Integrated with Lux GPT and Open Terminal on the same control plane so chat, execution, and wiki updates share one security model
Results
✓Enabled governed document-to-wiki automation with semantic diff, auto-publish, and reviewer-approved publishing fallback
✓Connected agentic knowledge to the KC Enterprise AI platform through shared LiteLLM integration
✓Improved retail IT knowledge discoverability for POS operations with MCP-accessible wiki content