KGMD

KGMD is an AI-native knowledge governance platform that transforms organizational knowledge into a trusted knowledge graph for humans and AI agents.

Comment

KGMD

Inspiration

As organizations adopt AI, one challenge is becoming increasingly clear: AI is only as reliable as the knowledge it receives.

Organizational knowledge is often scattered across Markdown files, documents, spreadsheets, databases, APIs, and internal systems. Even when the information exists, it may be duplicated, inconsistent, outdated, or missing the context and governance AI needs to use it confidently.

We wanted to solve that problem.

Instead of building another knowledge base, we envisioned a governed source of truth designed for both people and AI agents. That vision became KGMD.

KGMD began as a Knowledge Graph Markdown Database. It is now evolving into an AI-native Knowledge Governance Platform.

What it does

KGMD organizes an organization’s canonical knowledge into a persistent, governed workspace that humans and AI agents can use together.

In the working Build Week implementation, organizations can manage canonical Markdown documents through KGMD’s protected web application. ChatGPT connects to the deployed KGMD service through Model Context Protocol (MCP) tools that can:

  • Search governed organizational knowledge
  • Browse canonical folders
  • Read individual knowledge pages
  • Create and update pages
  • Move content within the Canon
  • Manage page attachments
  • Preserve knowledge outside an individual AI conversation

This gives ChatGPT access to persistent organizational knowledge without forcing the organization to place its entire history into every prompt.

KGMD’s broader semantic architecture brings together:

  • Knowledge Governance
  • Canonical Knowledge
  • Ontologies and Taxonomies
  • Knowledge Graphs
  • Entities, Attributes, and Relationships
  • Evidence and Provenance
  • Version History
  • Schema.org Mappings
  • Human Approval Workflows
  • Model Context Protocol Integration

Instead of simply storing information, KGMD turns disconnected organizational knowledge into structured, governed intelligence.

How we built it

KGMD was designed from the ground up as an AI-native platform.

The application uses:

  • Python
  • FastAPI
  • MySQL
  • Docker
  • REST APIs
  • OAuth
  • Markdown-based canonical documents
  • Model Context Protocol

The protected KGMD workspace allows people to browse, search, create, and maintain organizational knowledge. The backend stores the Canon and exposes controlled operations through its API and MCP service.

The KGMD MCP integration securely connects ChatGPT to the platform. Once connected, ChatGPT can search, retrieve, create, and update governed knowledge through defined tools instead of relying only on temporary conversation context.

We used Codex throughout the project to:

  • Inspect and understand the existing codebase
  • Design and implement application features
  • Develop and validate database migrations
  • Debug authentication, OAuth, and MCP integration
  • Deploy KGMD into its production infrastructure
  • Test the complete ChatGPT-to-KGMD workflow
  • Refine the platform’s semantic and governance architecture

GPT-5.6 helped us reason across the application, database, authentication, infrastructure, and knowledge architecture as one connected system.

This created a working bridge between AI reasoning and persistent organizational knowledge.

Challenges we ran into

The biggest challenge was not simply writing code. It was defining knowledge.

Organizations use different terminology for the same concepts. They may have duplicate documents, inconsistent taxonomies, undocumented relationships, and conflicting versions of what should be considered authoritative.

We had to think beyond traditional folders and files and begin defining:

  • What makes knowledge canonical
  • How knowledge should be organized
  • How relationships between concepts should be represented
  • How AI agents should discover and use trusted information
  • Which operations AI may perform
  • Where human governance and approval should remain

Another major challenge was securely connecting ChatGPT to KGMD. We had to implement and debug the application, database, OAuth flow, MCP service, deployment environment, and public endpoint before the complete integration worked.

The result is a system designed to serve humans and AI agents without sacrificing persistence, control, or transparency.

Accomplishments that we're proud of

  • Built and deployed a working AI-native knowledge platform
  • Created a persistent organizational Canon based on Markdown
  • Developed a protected workspace for managing governed knowledge
  • Connected ChatGPT directly to KGMD through MCP
  • Implemented tools for searching, reading, creating, updating, moving, and managing canonical content
  • Implemented OAuth-based access for the KGMD MCP service
  • Established the semantic foundation for ontologies, taxonomies, entities, relationships, evidence, and provenance
  • Demonstrated that an AI agent can work with organizational knowledge outside its temporary conversation context
  • Used Codex and GPT-5.6 across development, debugging, architecture, deployment, and verification

Our proudest accomplishment is that KGMD is not only an idea or an architecture diagram.

It is deployed, connected to ChatGPT, and working.

What we learned

AI does not simply need more data.

AI needs better knowledge.

We learned that persistent, governed knowledge can give AI systems greater consistency, context, traceability, and usefulness.

We also learned that connecting an AI agent to organizational knowledge requires more than search. The knowledge must have structure, ownership, history, permissions, and a clear definition of what should be trusted.

Codex accelerated the work far beyond code generation. It helped us reason across multiple systems, diagnose integration failures, validate deployments, and maintain continuity while the platform evolved.

The next generation of AI applications will need more than models and prompts. They will need a trusted knowledge layer.

What's next for KGMD

Our vision is to make KGMD the trusted knowledge layer for AI-powered organizations.

The next phase will expand KGMD with:

  • AI-assisted knowledge discovery and classification
  • Deeper ontology and taxonomy management
  • Entity and relationship exploration
  • Evidence and provenance workflows
  • Conflict and duplication detection
  • Knowledge quality and confidence scoring
  • Schema.org and external standards mapping
  • Enterprise integrations
  • Expanded approval and governance workflows
  • Collaboration among multiple AI agents
  • Controlled knowledge delivery to different models and applications

As AI agents become more capable and autonomous, organizations will need a reliable way to determine what those agents know, where that knowledge came from, and whether it can be trusted.

We believe governed knowledge will become as fundamental to AI-powered organizations as databases are today.

Built With

Share this project:

Updates

posted an update —

Post-submission update — August 14, 2026

KGMD didn't stop evolving when we submitted it.

Since the Build Week deadline, we've continued using KGMD in production, and something interesting happened: the system began validating the thesis behind the project.

KGMD has evolved beyond a Markdown knowledge graph into a governed knowledge and operational control layer supporting multiple real systems. We've added deterministic desired-state reconstruction, governed objects and relationships, provenance and recovery history, verification workflows, and increasingly agent-driven operations.

Perhaps the biggest change is how we're using it: KGMD is now helping govern the development of the systems that depend on KGMD.

The original submission said our proudest accomplishment was that KGMD was deployed, connected to ChatGPT, and working.

Today, we'd add one more sentence:

Then we started running the company through it.

The original submission remains the snapshot of KGMD at the Build Week deadline. We'll use these updates to document what happens as we continue putting the architecture into real-world operation.

Log in or sign up for Devpost to join the conversation.