Agent-PKM
Agent-optimized knowledge base derived from my Obsidian vault — journals, highlights, people, and goals as structured, queryable data for AI agents.
Agent-PKM watches my Obsidian vault and transforms a messy, interconnected personal knowledge system into structured data that AI agents can actually use — deep context about my goals, projects, people, and learned wisdom, without an agent having to parse a decade of free-form markdown.
What It Produces
- JSONL indices — one JSON object per line, grep-friendly for fast lookups
- YAML profiles — structured entity data for people, projects, and life areas
- Markdown prose — full journal text and resource details where nuance matters
- Graph edges — typed relationships between entities (friend_of, collaborates_on, aligns_with_goal)
- Timeline metrics — daily mood and energy trends, plus goal progress signals
The Scale
A year of daily journals with AI-extracted mood, energy, topics, and decisions. Over 4,300 highlight chunks from 460 books, articles, podcasts, and videos. More than a hundred people records, twenty projects, and a typed knowledge graph connecting all of it — fronted by an MCP server on the homelab so any agent (Ping, Claude Code, Zarq workers) can query it.
Why It Matters
Every AI project I build gets smarter because this exists. When Ping answers a question about a friend, when Pulse notices I’m sliding on a goal, when a Zarq agent needs to know what I actually care about — they’re all reading from the same agent-readable version of my brain.