Mond(AI)y Coffee

What are you building?

An Obsidian-first memory system for AI tools. A structured folder, a startup ritual, and a handful of command phrases that give Claude, Codex, or any agent persistent context — so every session can pick up where the last one left off, without re-explaining the project from scratch.

The core idea

AI tools are stateless by default. Every session starts at zero. The fix isn’t a new app or a vector database — it’s a well-organized folder and a consistent startup sequence.

The system lives in AI/ inside my Obsidian vault. When I start a session, Claude reads six files in order:

1. START.md — orientation
2. Adam.md — who I am, how I work
3. Context/Preferences.md — style and working defaults
4. Context/TechStack.md — tools I actually use
5. Context/CurrentState.md — what I’m actively working on
6. The relevant Projects/[Project].md — full project state

After that, the agent knows enough to be useful without a five-minute briefing.

What’s working / not working

Working:

• Phase 1 is live — git-tracked AI/ folder with full folder structure, operating rules, and three seeded domain Brains (House, Car, HomeLab)
• _System/AI-Behavior.md is the bootstrap file — Claude reads it on every session start
• INDEX.md is the machine-maintained manifest of all projects and Brains
• Session closeout is a command: say “Capture this session” → 8-step workflow writes updates back to Obsidian

Not working yet:

• Bridge files for active code projects (AGENTS.md / CLAUDE.md pointers) aren’t in place — agents launched inside a project folder don’t yet auto-discover the AI brain
• Placeholder content in Brains/House/STATE.md and Brains/Car/STATE.md
• No remote git backup yet

Open questions

• What’s the right threshold for “this project needs a bridge file”? Every code repo? Only active ones?
• Should people, contacts, or relationships get their own Brain section eventually?
• How much maintenance is too much — where does “useful system” become “second job”?

Spicy takes

• Most “AI memory” products are the wrong abstraction. You don’t need a proprietary tool or a hosted embedding store. You need a folder, a checklist, and discipline.
• The startup ritual matters more than the folder structure. Six files read in order beats a thousand disorganized notes.
• Plain markdown is the right format — readable by humans, readable by any AI tool, version-controllable, and zero vendor lock-in.
• Context is a habit, not a feature. The system only works if you close sessions properly. “Capture this session” has to become as automatic as saving a file.

If we only discuss one thing

What would your startup ritual look like?

The six-file sequence is the kernel of the whole system. Everything else — folder structure, Brains, commands — is scaffolding around that ritual. If you wanted to steal one thing and adapt it to your own setup, that’s the one.

Built on: Obsidian + plain markdown + git. No special tools required.