Skip to content

OpenWiki and FastMCP architecture

An in-depth explanation of the technical design, data flows, and reasoning behind OpenWiki and FastMCP integrations.

High-level architecture

The DSOM knowledge layer bridges localised documents to active AI agents via two complementary components:

flowchart TD
    Files["Sovereign Markdown Files<br/>(docs/, openwiki/)"] -->|Local Indexing| Server["FastMCP Server<br/>(tools/mcp/server.py)"]
    Files -->|Compilation| Web["Static Docs Site<br/>(GitHub Pages / RTD)"]
    Server -->|JSON-RPC via stdio| Agent["AI Agent<br/>(Cursor, Copilot, Claude)"]

Why pure Python OpenWiki?

Historically, OpenWiki implementations required heavy Node.js runtimes or cloud compiler keys. This design violated Sovereign AI Rule 27 (Zero-Binary Mandate).

Replacing Node binaries with the pure Python openwiki_emulator.py provides: - Zero Node.js dependencies: The script runs offline using standard Python library APIs and the pyyaml package. - In-place self-healing: The emulator scans embedded diagrams, detects syntactical errors, degrades blocks gracefully, and heals them when fixed.

FastMCP context routing

The FastMCP server (tools/mcp/server.py) acts as the primary API broker for AI engines.

Rather than feeding the entire documentation codebase into the context window, the server enables Progressive Disclosure: 1. The AI reads condensed states (dsom://brain/state). 2. If further details are needed, the AI calls targeted tools (search_palace or search_openwiki). 3. Only the relevant context fragments are returned, reducing overall token usage.

Deep State of Mind (DSOM) For My AI Protocol | Harisfazillah Jamel (LinuxMalaysia) | 2026-08-14 Standard: UK English | DBP-standard Bahasa Melayu Malaysia (Piawai) | GNU General Public License v3.0