MyKG: Universal Knowledge Graph

An LLM-maintained personal knowledge base. The LLM incrementally builds and maintains a structured wiki of markdown files — reading sources, extracting knowledge, updating pages, and keeping everything connected.

Setup

1. Initialize

# From kgskills repo
cp -r templates/kg-init/* ~/MyKG
# Or use /kg-init command

2. Configure Path

# Option A: Environment variable (add to ~/.bashrc or ~/.zshrc)
export MYKG_PATH="$HOME/MyKG"

# Option B: Config file
mkdir -p ~/.config/mykg
echo "/path/to/your/kg" > ~/.config/mykg/config

3. (Optional) Use with Obsidian

Point your Obsidian vault at the MYKG_PATH for graph view, backlinks, and browsing.

Optional Skill Integrations

These companion skills from kepano/obsidian-skills enhance mykg when installed.

If these skills are loaded, prefer them over raw file edits or WebFetch for the relevant operations.

KG Path Resolution

The skill resolves the KG path in this order:

1. MYKG_PATH environment variable
2. ~/.config/mykg/config file
3. ~/.mykgrc file
4. Common locations: ~/MyKG, ~/Dev/Obsidian/MyKG
5. Ask user for location

Three Layers

┌─────────────────────────────────────────────┐
│              Raw Sources (immutable)         │
│  Papers, articles, transcripts, notes, URLs │
└──────────────────────┬──────────────────────┘
                       │ ingest
                       ▼
┌─────────────────────────────────────────────┐
│              Wiki (LLM-maintained)           │
│  Summaries, entity pages, concepts,         │
│  synthesis, cross-references                │
└──────────────────────┬──────────────────────┘
                       │ query
                       ▼
┌─────────────────────────────────────────────┐
│              Schema (configuration)          │
│  AGENTS.md — tells the LLM how to maintain  │
└─────────────────────────────────────────────┘

Raw Sources (raw/)

Immutable source documents. The LLM reads from them but never modifies them. This is the source of truth.

Wiki (wiki/)

LLM-generated markdown files. Summaries, entity pages, concept pages, comparisons, synthesis. The LLM owns this layer — it creates pages, updates them, maintains cross-references, and keeps everything consistent. You read it; the LLM writes it.

Schema (AGENTS.md)

Tells the LLM how the wiki is structured, what conventions to follow, and what workflows to use. Co-evolve this over time.

Structure

MyKG/
├── index.md              ← Catalog of all pages. START HERE.
├── log.md                ← Chronological change history.
├── AGENTS.md             ← Schema: structure, conventions, workflows.
├── Active.md             ← Current focus items and priorities.
├── raw/                  ← Source documents (immutable).
│   ├── articles/
│   ├── papers/
│   ├── transcripts/
│   └── notes/
├── wiki/                 ← LLM-generated wiki pages.
│   ├── entities/         ← Pages for specific entities (people, orgs, tools).
│   ├── concepts/         ← Pages for ideas, topics, themes.
│   ├── synthesis/        ← High-level summaries and overviews.
│   └── comparisons/      ← Side-by-side analyses.
└── <Project>/            ← One subdirectory per project.
    ├── SKILL.md          ← Project-specific skill (optional).
    ├── AGENTS.md         ← Project-specific schema (optional).
    ├── raw/              ← Project sources.
    └── wiki/             ← Project wiki pages.

Operations

Ingest

Read a source, extract knowledge, integrate into the wiki:

1. Read the source document
2. Discuss key takeaways with user
3. Write a summary page in wiki/
4. Create/update entity pages in wiki/entities/
5. Create/update concept pages in wiki/concepts/
6. Note contradictions with existing pages
7. Update index.md with new pages
8. Append entry to log.md

A single source might touch 10-15 wiki pages. The knowledge compounds.

Query

Search the wiki, synthesize an answer with citations:

1. Read index.md to find relevant pages
2. Drill into specific pages
3. Synthesize an answer
4. Good answers can be filed back as new wiki pages — comparisons, analyses, connections discovered during exploration. This way your queries compound too.

Lint

Periodically health-check the wiki:

• Orphan pages (no inbound links)
• Stale claims superseded by newer sources
• Important concepts mentioned but lacking their own page
• Missing cross-references
• Contradictions between pages
• Data gaps that could be filled with a web search

Two-Tier Skill System

Universal Skill (this file)

• KG structure and conventions
• Navigation and search workflows
• Standard operations (ingest, query, lint)
• Cross-project references

Project Skills (<Project>/SKILL.md)

• Domain-specific page types and conventions
• Custom ingestion workflows
• Domain terminology and knowledge
• Project-specific formatting rules

When working on a project, both skills are available:

Universal SKILL (always loaded)
    ↓
Project SKILL (loaded when working on that project)
    ↓
Wiki Content (entities, concepts, synthesis)

When to Use

Use when:

1. Ingesting a new source → extract and integrate
2. Answering questions → search wiki first
3. Exploring a topic → read wiki pages, follow links
4. Maintaining the wiki → lint, update, cross-reference
5. Starting a new domain → create project with custom SKILL.md

Don't use for:

• Code implementation details (project AGENTS.md)
• Ephemeral conversation context (chat history)

Workflow

Ingesting a Source

1. Place source in raw/ (or provide URL/file)
2. Read the source, discuss key takeaways
3. Create/update wiki pages (entities, concepts, synthesis)
4. Update index.md
5. Append to log.md

Querying the Wiki

1. Read index.md to find relevant pages
2. Read specific pages
3. Synthesize answer with citations ([[Page Name]])
4. Optionally save answer as new wiki page

Creating a New Project

1. Copy project template: cp -r project-templates/init "$KG_PATH/ProjectName"
2. Edit <Project>/SKILL.md with domain-specific instructions
3. Edit <Project>/AGENTS.md with project schema
4. Update root index.md
5. Append to log.md

Linting the Wiki

1. Check for orphan pages
2. Find contradictions between pages
3. Identify missing cross-references
4. Flag stale claims
5. Suggest new sources to investigate

Page Types

Conventions

Wikilinks

• [[Page Name]] - internal link
• [[Page Name#Section]] - deep link
• [[Page Name|Display Text]] - alias
• [[Project/wiki/Page]] - cross-project link

Frontmatter

---
tags:
  - category
  - subcategory
created: YYYY-MM-DD
updated: YYYY-MM-DD
sources:
  - raw/article-name.md
related:
  - [[Related Page]]
---

File Naming

• Lowercase with hyphens: entity-name.md
• Wiki pages in appropriate subdirectory
• Raw sources preserve original name or slug

Status Values

• draft - Work in progress
• in-progress - Actively being updated
• complete - Finished

Key Files

Index Format

index.md is content-oriented. Catalog of everything with links and one-line summaries. Organized by category. The LLM reads it first to find relevant pages.

## Entities
| Page | Summary | Sources |
|------|---------|---------|
| [[entities/some-person]] | Researcher specializing in X | 3 |

## Concepts
| Page | Summary | Sources |
|------|---------|---------|
| [[concept/topic]] | Overview of topic Y | 5 |

Log Format

log.md is chronological. Append-only record of what happened. Consistent prefixes make it parseable:

## [2026-04-07] ingest | Article Title
- Added entity pages: X, Y
- Updated concept: Z
- New contradiction noted: A vs B

## [2026-04-06] query | What is X?
- Answer filed as [[concepts/X explained]]

Parse with: grep "^## \[" log.md | tail -5

Cross-Project References

Projects can reference each other:

Related: [[OtherProject/wiki/concepts/Topic]]
Contradicts: [[OtherProject/wiki/entities/Entity]]

Maintenance

After each session:

1. Update log.md with changes
2. Update index.md for new pages
3. Update Active.md if priorities changed

Periodically:

1. Run /kg-lint to find orphans and contradictions
2. Review synthesis pages for completeness
3. Archive stale content