Agentic Memory Template

A beginner-friendly template for building a governed AI workspace.

This architecture is based on: Beyond Karpathy's LLM-Wiki: The Necessity of Cognitive Governance

Start in 60 seconds

You do not need to code to use this.

Option A: Download ZIP

1. Click Code -> Download ZIP.
2. Unzip the folder.
3. Open the folder in your AI tool (Claude, Cursor, Codex, etc.).
4. Send this command:

Read START-HERE.md and run the setup interview now.

Option B: Clone with Git

git clone https://github.com/jonadas-tech/agentic-memory-template.git
cd agentic-memory-template

Then open the folder in your AI tool and send:

Read START-HERE.md and run the setup interview now.

What happens after this command

The AI asks a short onboarding interview (about 5-10 minutes) and personalizes:

• AGENTS.md (how the AI should work with you)
• STYLE.md (your voice and writing preferences)
• INDEX.md (what the AI should read first for each task)
• TASKS.md (your active work)

Then it proposes your first working session.

Before and after (real difference)

Ungoverned output usually sounds smooth but generic:

"This source discusses leadership strategy and long-term thinking."

Governed output forces structure, tension, and reusable links:

"This claim reframes 'strategy' as rivalry regulation, not optimization.
Antagonist: infinite-game rhetoric without mimetic analysis.
Related: LIT-Sinek, ZET-Mimetic-Rivalry."

This template is built to produce the second type by default.

Your first 10 minutes (first win)

1. Run setup with: Read SETUP.md and run the full setup interview now.
2. Put one real note in content/inbox/.
3. Run: Run ingest-protocol for the latest inbox source.
4. Confirm you got:
   • 1 LIT note
   • 1 ZET note (or explicit reason not to create one)
   • 1 MOC update
5. End with: Run session review.

If you complete this loop once, the system is working.

Why this template works

Most AI workflows optimize for smooth language. This template optimizes for compounding thinking.

The core pattern:

1. Keep raw material immutable (content/inbox/).
2. Use governance rules (SCHEMA.md).
3. Compile into linked notes (docs/kb/).
4. Run short session reviews so quality improves over time.

Included

• START-HERE.md: one-page onboarding command center
• SETUP.md: interview script the AI follows to personalize your workspace
• SCHEMA.md: cognitive governance rules
• SESSION-REVIEW.md: end-of-session protocol
• .claude/skills/ingest-protocol/SKILL.md: ingest skill instructions
• worked ingest example (source -> LIT -> ZET -> MOC)

Directory map

• content/inbox/: raw sources (read-only for AI)
• docs/kb/01-permanent-notes/: ZET notes
• docs/kb/02-literature-notes/: LIT notes
• docs/kb/03-maps-of-content/: MOC notes
• memory/: project memory and session logs
• workflows/: guided recurring processes

Suggested first week

1. Run onboarding (START-HERE.md command).
2. Ingest 1-3 real sources into content/inbox/.
3. Run one drafting session using your new files.
4. Close each session with: Run session review.

If you are advanced

You can skip onboarding and edit files directly.

Start with:

1. SCHEMA.md
2. AGENTS.md
3. INDEX.md
4. STYLE.md