Agentic Cortex: Using Dendron's naming hierarchy as an AI agent's knowledge graph

Hi Dendron community!

I wanted to share a project that builds directly on Dendron's hierarchical naming convention to create a personal AI operating system.

The core insight

Dendron's dot-notation naming (user.*.md, proj.2026.*.md, meet.2026.03.*.md) naturally forms a queryable knowledge graph. An AI agent can navigate the entire vault using glob patterns and grep -- no vector database, no embeddings, no special infrastructure. The filename IS the schema.

A real vault's note graph. Each cluster is a Dendron namespace.

What the system does

• Morning briefings from vault state (calendar, stale follow-ups, project momentum)
• Email drafting in your voice (extracted from sent emails into a VOICE_PROFILE.md)
• Ambient context from Screenpipe (local screen/audio recording → daily journals)
• Structured feedback that compounds -- correct the agent once, it saves the rule + rationale + scope, behavior changes permanently
• People graph -- 400+ person profiles as connective tissue across all data sources

The AI reads the vault at session start, syncs email/calendar/screenpipe data into vault notes via MCP, and maintains a three-tier memory system (working memory / context model / vault) -- all in plaintext markdown following Dendron conventions.

Links

• GitHub: https://github.com/Albert-Ying/agentic-cortex
• Blog post: https://kejunying.com/blog/agentic-cortex/

The repo includes a seed vault, skills, a setup script, and a 9-chapter tutorial. Built for OpenClaw / Claude Code but the architecture works with any agent that can read markdown.

Would love to hear if anyone else is using their Dendron vault as an AI memory layer!