nduckmink/arkon

What it solves

Arkon is designed to prevent fragmented AI adoption in organizations where employees manually copy-paste internal documents into chatbots, leading to inconsistent context and security risks. It provides a centralized, self-hosted knowledge management layer that ensures employees receive the correct, permission-scoped internal data when interacting with AI clients like Claude.

How it works

Arkon acts as an MCP (Model Context Protocol) server that compiles organizational data (SOPs, policies, and internal docs) into a structured, traceable knowledge wiki. It uses a specialized MRP pipeline (Map, Reduce, Plan-review, Refine, Verify, Commit) to transform raw documents into interlinked wiki pages rather than just chunking them into a vector database. This wiki is then served to LLMs via a single endpoint, with access filtered by the user's department and role.

Who it’s for

Enterprise teams and organizations that need a secure, self-hosted way to manage internal knowledge for AI agents and employees, specifically those using Claude Desktop or Claude.ai.

Highlights

• MRP Pipeline: Compiles documents into a coherent wiki with human-reviewable plans, page merging to prevent knowledge loss, and traceable claims.
• MCP Server Integration: Connects directly to AI clients via OAuth 2.1 + PKCE, providing tools for wiki search, source drill-down, and edit workflows.
• Departmental Scoping: Enforces hard boundaries between departments (e.g., HR, Legal, Engineering) and a global company-wide scope.
• Fine-Grained RBAC: Role-based access control with roles like Viewer, Contributor, Editor, and Admin, including an audit log for privileged actions.
• Pluggable AI Providers: Supports a curated list of LLMs (Anthropic, Google, OpenAI) and embedding models with the ability to perform online re-embed migrations.
• Self-Hosted & Private: Deployed via Docker with no telemetry and encrypted API keys.