Dennis OS: Shared Context for AI Agents

The operating system behind everything on this site. Humans browse it in Obsidian for AI agents; the named AI agents work in it directly. This page is the map.

Dennis OS is a folder of plain-text Markdown files — an Obsidian vault — that functions as my second brain and my agents’ first brain. It holds who I am, what I believe, every meaningful decision, my frameworks, my offers, my relationships, live project state, and a signed log of every AI work session. It is the reason my AI output compounds instead of resetting every conversation.

Dennis Yu keynoting on the main stage at the National Funeral Directors Association conference, presenting the four stages of the Content Factory on a large screen
Dennis Yu presents the four-stage Content Factory at the National Funeral Directors Association conference—the human operating model Dennis OS turns into reusable context and agent workflows.

Dennis OS — raw material in, durable leverage out 1 · CAPTURE Raw notes, transcripts, calls, ideas, pasted AI memory packets — messy on purpose. inbox/ 2 · DISTILL Who Dennis is, what he believes, decisions made, live state, model routing. memory/ · decisions/ · STATE 3 · LEVERAGE Named frameworks, offers, audiences, relationships — the teachable, sellable core. frameworks/ · offers/ 4 · SHIP Articles, tools, skills, products, meta-articles — published proof. outputs/ Worked at every stage by four AI assistants under one protocol (AGENTS.md) Claude ChatGPT · Codex Gemini Grok Every session: boot from state → claim → work → signed log → update. Humans and AIs share the same map. Stage colors are one blue ramp — one pipeline, not four products. Assistant colors mark identity and are always labeled.

The pipeline in plain words

Raw material goes in messy — meeting transcripts, ideas, pasted AI session packets. Agents distill it into durable memory: canonical facts, principles, decisions with reasoning attached. That memory feeds the leverage layer — the named frameworks I teach (Dollar a Day, Content Factory, and the rest) and the offers built on them. And the leverage layer ships: articles, tools, skills, products — each with a meta-article documenting how it was made, per The System.

The operating rule: capture first, distill second, package third. No note has to be perfect on entry.

What makes it different: the AIs share it

Since July 2026, Dennis OS doubles as cross-vendor shared memory. Claude, ChatGPT/Codex, Gemini, and Grok all boot from the same live state file, claim their work so they don’t collide, and sign their session logs with their exact model name. Two of them — Claude and Codex — built that memory layer together in one night, including a real mid-build collision and merge you can read in the git history. The context migration workflow is the fast on-ramp when the useful history is still trapped in one chat.

📖 The story (meta-article)

How I Built Shared AI Memory Without Lock-In — two frontier models, one folder, receipts included.

🔧 The recipe (Definitive Article)

Set Up Cross-Agent Shared Memory — 15-minute setup, starter files, and an assignment written directly to your AI.

🤖 For AI assistants

Working with Dennis? Boot from the public agent context capsule — safe context plus the protocol, at one stable URL.

How it connects to everything else

Dennis OS is the memory layer of The System, my public AI-marketing machine. The Task Library holds 1,000+ cataloged tasks; each task’s Definitive Article is the SOP; every execution produces a meta-article; and the agents that do the work — including the two memory skills, the Memory Keeper and the Memory Librarian — are cataloged on the Content Factory floor. Meta-articles feed results back into the definitive articles, the OS captures the decisions, and the loop compounds. That’s the whole trick: nothing my agents learn is allowed to evaporate.

Want this for yourself? Start with the Definitive Article — or just show it to your AI and say “set this up for me.”

Maintained by Dennis’s agents under the Dennis OS write-back protocol. Last updated July 22, 2026 by claude-fable-5.