Concept Presentation

Markdown as an
Operating System
for Enterprises

All processes, tasks, decisions, and knowledge as Markdown files — so every employee can work with AI agent tools.

10
Concept docs
40+
Example files
30+
Compatible tools
5
Interfaces
scroll

00 — Overview

The idea in one sentence

All enterprise processes, tasks, decisions, and knowledge live as Markdown files in a Git repository — so every employee can work directly with AI agent tools (Kivanto, Claude Code, Codex CLI, Copilot CLI, Gemini CLI, and more).

The triangle that makes it work

Versioning Context Universal Format AI Agents 30+ tools Git Tracking Markdown Files Company Knowledge

Who is this for?

SizeEffortBenefit
3–10 people1 day setupCentral knowledge base, AI as "all-knowing colleague"
10–50 people1–2 weeksUnified processes, review workflows, onboarding acceleration
50–500 people2–3 monthsCross-department transparency, compliance, AI self-service

Four Layers

Layer 4: Agent Interface AGENTS.md / SKILL.md / Natural Language Layer 3: Knowledge Graph Entities, Relations, Impact Analysis Layer 2: RAG Pipeline Chunking, Embeddings, Hybrid Search, Freshness Layer 1: Markdown Filesystem Files + YAML Frontmatter + Git

Why Markdown?

Human-readable and machine-readableNo special tools needed
Git-nativeFull audit trail, branching, review workflows
Agent-agnosticEvery AI tool already understands Markdown
No vendor lock-inPlain text files, no proprietary format

01 — Collaboration

Git as the foundation

Core decision: Git as the foundation, with bridges for non-technical users.

Git vs. Cloud Sync

ProblemCloud SyncGit
Simultaneous editingCreates duplicatesLine-based merging
Change trackingFile versions onlyLine-by-line diff
UndoWhole file rollbackIndividual changes
Review before mergeNot possiblePull requests

3-Tier Model

Tier 1: Technical
Terminal + AI Agent
VS Code + Extensions
IT, Engineering, DevOps
Tier 2: Semi-technical
Obsidian + Git Plugin
Kivanto App
Project Managers, Team Leads
Tier 3: Non-technical
GitHub.dev (Browser)
GitBook / Decap CMS
HR, Marketing, Finance

Conflict prevention

02 — Folder Structure

One repo, clear hierarchy

company-repo/ AGENTS.mdAI agent context (single source of truth) company/Vision, strategy, org chart, glossary processes/Cross-department (onboarding, procurement) departments/Per department: processes/ templates/ policies/ hr/processes/ templates/ policies/ skills/ finance/processes/ policies/ skills/ product/concepts/ decisions/ skills/ marketing/processes/ brand/ skills/ engineering/processes/ decisions/ skills/ projects/Time-limited, cross-team meetings/Minutes (chronological by year) results/AI agent outputs, analyses, reports skills/Company-wide agent skills

YAML Frontmatter — every file's metadata

Frontmatter schema
---
title: Recruiting Process          # required
owner: priya-sharma                 # required, firstname.lastname
department: hr                      # optional, enum
type: process                       # required, enum
status: active                      # required, enum
created: 2024-08-10                 # required, YYYY-MM-DD
last-reviewed: 2026-03-25          # optional, used for freshness
tags: [recruiting, hiring]          # optional
created-by: claude-code             # optional, marks AI-generated
---
FieldPurpose
ownerWho is responsible? Who reviews changes?
departmentAI agent can narrow context
typeAutomatic aggregation (e.g., list all policies)
statusDrafts ≠ valid, archived ≠ active
last-reviewedFreshness check: warning at >6 months
created-byMarks AI-generated content

03 — AI Agent Integration

Context file hierarchy

company/AGENTS.md "We are TechVision, 80 employees, SaaS. Frontmatter schema: ..." inherits departments/hr/AGENTS.md "HR team, 5 people. GDPR rules. Priority: Onboarding revamp." AI agent understands: Company + Department + Rules

Where do AI results go?

What the agent createsStored in
New processdepartments/<dept>/processes/
Meeting minutesmeetings/YYYY/
Analysis / reportresults/
Change to existing docIn-place (same file)
Decision documentationdecisions/

5 concrete scenarios

1
HR creates process — AI reads existing processes as reference, creates new file with frontmatter
2
Team lead searches info — AI finds procurement process + budget limits cross-department
3
Meeting processed — Notes → minutes + extracted action items with deadlines
4
Report generated — Policy overview with freshness status → results/
5
New employee — Personalized onboarding from vision + org chart + processes

04 — Change Tracking & Quality

Automated quality gates

CI/CD Pipeline

markdownlint Frontmatter Link Check ✓ Merge Syntax Required fields References All OK!

Freshness monitoring

StatusAgeAction
Current<4 monthsNo action needed
Due soon4–6 monthsPlan review
Overdue>6 monthsImmediate review by owner

Review workflow

Change typeReview needed?Reviewer
Typos, formattingNo—
Content additionYesFile owner
Process changeYesDepartment lead + owner
Cross-departmentYesAll affected owners

05 — Rollout Guide

3-phase introduction

Phase 1: Foundation

1–2 weeks. Repo setup, AGENTS.md, CI/CD, core team (2–3 people).

Phase 2: Pilot

2–4 weeks. One department fully migrated. Feedback collected.

Phase 3: Rollout

4–8 weeks. More departments. Non-technical users. Old system off.

Don't copy 1:1. Migration is the chance to clean up — don't migrate outdated docs, merge redundant content, adapt to the new structure.

Migration paths

SourceMigration path
Confluence / WikiExport as HTML → convert with AI agent → review
Word / Google DocsAI agent reads directly, or Pandoc batch conversion
NotionMarkdown export → clean up → add frontmatter

06 — Tool Comparison

Editor matrix

ToolTierGitWYSIWYGOfflineAI Agent
VS Code + CC1–2UIPreviewYesYes
Obsidian + Git2PluginPreviewYes*
Kivanto App2–3Built-inYesYesYes
GitHub.dev3Built-inPreviewNoNo
GitBook3SyncYesNoNo
Decap CMS3DirectFormsNoNo

* Obsidian + AI agent CLI in the same folder

Decision tree

Technically proficient? Yes VS Code + AI Agent No Edits regularly? Yes Desktop app possible? Obsidian + Git / Kivanto App GitBook / Decap CMS No GitHub.dev

07 — Tool Agnosticism

AGENTS.md as open standard

AGENTS.md is the open standard for AI agent context files — managed under the Linux Foundation, supported by 30+ tools.

This system is model-agnostic: it is independent of any specific AI model — whether Claude, GPT, Gemini, Llama, Mistral, or any future model. The knowledge base is plain Markdown and Git, with no dependency on a particular vendor’s AI.

The tool landscape

ToolMakerTypeOpen Source
KivantoKivanto.aiCLI + Desktop + WebMIT
Claude CodeAnthropicCLI + IDENo
Codex CLIOpenAICLIYes
Copilot CLIGitHubCLI + IDENo
Gemini CLIGoogleCLIYes
CursorAnysphereIDENo
AiderOpen SourceCLIYes

Reference strategy: no duplicate content

AGENTS.md Full content (single source) CLAUDE.md "Read AGENTS.md" .kivanto/KIVANTO.md "Read AGENTS.md"

One file to maintain. Full tool compatibility.

08 — Skills

Agent Skills standard

Agent Skills standard (agentskills.io) — supported by 30+ tools including Claude Code, Copilot, Cursor, Gemini CLI, Codex CLI.

One skill = one folder

Skill folder structure
meeting-minutes/
├── SKILL.md              ← Metadata + instructions
├── scripts/              ← Executable code (optional)
├── references/           ← Additional docs (optional)
└── assets/               ← Templates, resources (optional)

Progressive disclosure

1

Discovery

Name + description only
~100 tokens / skill

At startup, all skills

2

Activation

Full SKILL.md loaded
<5000 tokens

When skill matches

3

Execution

scripts/ references/
On demand only

Only when needed

Example skills

meeting-minutes
All departments
Create minutes from notes
freshness-check
All departments
Check document freshness
onboarding-summary
All departments
Briefing for new employees
job-posting
HR
Job posting from role description
adr
Product
Document a decision
blog-post
Marketing
Blog post per brand guidelines
incident-postmortem
Engineering
Blameless post-mortem

Skills vs. Templates vs. Processes

Template

Empty scaffold
Human fills in
employment-contract.md

Process

Step-by-step
Human follows
recruiting.md

Skill

AI instructions
Agent executes
job-posting/SKILL.md

09 — Example

TechVision GmbH

A complete implementation for a fictional SaaS company (~80 employees).

SaaS

FlowBoard

Project management SaaS product

~80

Employees

Munich (HQ) + Berlin (Engineering)

TS

Tech Stack

TypeScript, React, NestJS, PostgreSQL, AWS

40+

Files + 7 Skills

Realistic content, Agent Skills standard

What's in the repository

Company Vision & strategy, org chart, glossary
Processes Onboarding, travel expenses, procurement
HR Recruiting, performance review, remote work
Finance Invoice intake, spending limits, budgets
Product Roadmap 2026, ADR, AI assistant concept
Marketing Content approval, tone of voice
Engineering Deployment, release train, hotfix
Infra CI/CD, CODEOWNERS, JSON Schema

How an employee works with it

Terminal session
$ cd company-docs/departments/hr/

$ kivanto
  Welcome! I know the context of the HR department.

> Create a job posting for a Senior Backend Developer
  in the Engineering team.

  Using the skill "job-posting", reading the tone of voice
  guide and current benefits...

  ✅ Job posting created.

10 — Projects

mdaios & mdaifs

fw

mdaios — The Framework

The overarching concept: how to structure enterprise processes, tasks, and knowledge in Markdown. 10 concept documents, example repository, presentation, two academic papers.

fs

mdaifs — The Filesystem

The intelligent Markdown filesystem. Parses YAML frontmatter, builds a search index, auto-constructs a knowledge graph, and exposes everything through 5 interfaces.

mdaifs — 5 Interfaces

InterfaceDescription
CLI11 commands (search, graph, impact, freshness, validate, ...)
Python APIfrom mdaifs import MdaifsRepo
MCP Server7 tools for AI agents (Claude Code, Kivanto, etc.)
REST APIFastAPI with Swagger docs
FUSE MountVirtual filesystem with .mdaifs/ query directory
mdaifs CLI
$ pip install mdaifs
$ mdaifs status
$ mdaifs search "onboarding process"
$ mdaifs graph nina.scholz
$ mdaifs impact hr
mdaios on GitHub mdaifs on GitHub mdaifs.org

11 — Papers

Academic papers

P1

mdaios: Markdown-native Intelligent Filesystem for Human-AI Collaboration

The conceptual paper. 4-layer model, RAG on Markdown, automatic knowledge graph construction from frontmatter, freshness scoring, progressive disclosure for agents, comparison with existing systems.

P2

mdaifs: From Concept to Real System

The technical architecture paper. 5 implementation paths: FUSE virtual filesystem, system library (libmdaifs), MCP server, CLI, and REST API. Rust crate structure, migration strategy, performance targets.

Read Paper 1 (English) Read Paper 2 (English)

12 — Compatible Tools

Works with 30+ tools

mdaios works with any AI agent tool that reads Markdown.

ToolMakerType
KivantoKivanto.aiCLI + Desktop + Web
Claude CodeAnthropicCLI + IDE
Codex CLIOpenAICLI
Copilot CLIGitHubCLI + IDE
Gemini CLIGoogleCLI
CursorAnysphereIDE
AiderOpen SourceCLI
Roo CodeCommunityIDE
AmpSourcegraphIDE
Kilo CodeCommunityIDE
WindsurfCodeiumIDE
Continue.devOpen SourceIDE
ClineOpen SourceIDE
JunieJetBrainsIDE
GooseBlockCLI
OpenHandsOpen SourceCLI
New: Kivanto App — a free cross-platform desktop Markdown editor with built-in AI agent (30+ LLM providers), Git integration, and knowledge graph. Coming soon — available mid-April 2026.