Skip to content

MarcoBBo/hypomnema

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Hypomnema

Structured memory for AI-assisted work.

The ancient Greeks kept hypomnemata — structured notebooks of things read, heard, and thought, designed to be reread before acting. Not diaries. Not journals. Operational state, compressed for reuse.

2,400 years later, we need them again — for AI.


The Problem

AI assistants have no memory between sessions. Every new conversation starts from zero.

On simple tasks, this doesn't matter. On complex, multi-session projects — strategy, research, product development, investment analysis, writing — it's devastating. Decisions get re-debated. Context gets lost. The AI makes assumptions that contradict choices from three sessions ago. Documents pile up with no clear hierarchy. The project slowly drifts from its own history.

The common workaround — pasting large documents into context each time — is wasteful, noisy, and fragile. It treats the symptom, not the disease.

The disease is the absence of a structured practice for continuity.

The Solution

Hypomnema is a methodology, not a tool. It works with any AI (Claude, ChatGPT, Gemini, Copilot) and any document store (Notion, Google Docs, markdown files, a Git repo). No code. No plugins. No infrastructure.

The method has three components:

One document — the Session Brief — holds compressed project state: what's true now, what happened before, what to do next, every decision ever made, and where all documents live. One read restores full context.

Three rituals give the document its power:

Ritual When What happens
Resume Start of session AI reads the Session Brief. Announces the session. Starts working with full context.
Wrap Up End of session AI updates the Brief: current state, next task, decisions, documents. The next session gets a clean handoff.
Check Integrity Every ~10 sessions AI reads all project documents, cross-references against the Brief, reports drift. A full audit.

One home — the Project Hub — organizes everything the project produces. Every document is a child of the hub, registered in the Brief. Nothing gets orphaned. Nothing lives only in chat history.

That's it. One hub, three rituals, zero context loss.


Multi-Provider Continuity

This is Hypomnema's most powerful property, and it's inherent to the architecture — not an add-on.

The Session Brief is plain structured text in an external store. It contains no provider-specific syntax. Any AI that can read and write your document store can participate in the same project.

This means you can run Session 5 in Claude, Session 6 in Gemini, and Session 7 in ChatGPT — with perfect continuity. Each AI reads the same Brief, does its work, writes back to the same Brief. The project doesn't care which model is running.

When your document store is Notion, this works today across Claude (all products), ChatGPT (Business/Enterprise), Gemini CLI, Cursor, Windsurf, and any MCP-compatible environment. As MCP adoption grows, coverage expands automatically.

No other methodology or tool enables cross-provider project continuity.


Quick Start

1. Create a Project Hub

A parent page in your document store. Title, one-line description, list of documents.

2. Create a Session Brief

A child of the hub. Seed it with your current project state, first task, known constraints, and open questions. → Session Brief Template

3. Write Project Instructions

Paste the instructions template into your AI's custom instructions. Tell it where the Brief lives and how to run the three rituals. → Project Instructions Template

4. Start

First message: "Resume"

The AI reads the Brief, announces the session, and gets to work. When you're done: "Wrap up".

Full method guide · All templates · Project Hub template


Setup Prompts

Beyond the templates, Hypomnema includes ready-to-use prompts that automate the setup process:

Prompt What it does Runs in
Migration Prompt Scans an existing AI project's conversation history, extracts context, and builds the full Hypomnema infrastructure automatically. AI chat (e.g., Claude Projects)
New Project Prompt Sets up a new project from scratch — or imports a project from ChatGPT using a structured summary file. AI chat
ChatGPT Import Prompt Processes a ChatGPT data export: cleanses, classifies conversations by meaningfulness, and outputs organized files ready for import. Uses recursive synthesis for large projects (100+ conversations). Claude Code (needs filesystem access)

The prompts are self-contained — paste them into a chat and they guide you through the entire process.


Who Is This For?

Hypomnema is designed for knowledge workers using AI on complex, multi-session projects — not single-shot prompts or casual chat. If your AI work spans weeks or months, involves accumulated decisions, and produces documents that build on each other, this method eliminates the context tax you're paying today.

It works across any domain: software development, research, strategy, writing, investment analysis, design, legal work, consulting — any field where project continuity matters.

The method is human-driven by design. The human decides, the AI proposes and executes. There are no autonomous agents, no background processes, no infrastructure to maintain. You, a document, and a ritual.


What's Inside

hypomnema/
├── README.md                              # You are here
├── LICENSE                                # CC BY-NC 4.0
├── docs/
│   └── method.md                          # Full methodology guide
├── templates/
│   ├── project-hub.md                     # Project Hub template
│   ├── session-brief.md                   # Session Brief template
│   └── project-instructions.md            # AI custom instructions template
└── prompts/
    ├── migration-prompt.md                # Migrate an existing AI project to Hypomnema
    ├── new-project-prompt.md              # Set up a new project from scratch (or import from ChatGPT)
    └── chatgpt-import-prompt.md           # Process a ChatGPT export for import

Design Principles

Compressed history, not transcripts. Each session summary is one line. After 20 sessions you have 20 lines — not 20 conversations of tokens.

State over content. The Brief tells the AI what's true now, not how we got here. It's a pointer structure, not a copy.

Decisions are immutable. Never deleted, only superseded. The AI can reason over the evolution of thinking.

Cheap resume, expensive audit. Most sessions need one read. Integrity checks are rare and deliberate. Low per-session cost with a safety net for drift.

Human decides, AI proposes. The AI presents analysis and options first. The human approves. Then documents get updated. Never the reverse.


Origin

Hypomnema was developed by Marco Borza through extended use of AI on real projects across multiple domains. The patterns emerged from real friction — lost context, re-debated decisions, orphaned documents — and were refined into a generalizable methodology validated on projects ranging from strategy and investment to product development and research.

The name comes from the ancient Greek practice of hypomnemata (ὑπομνήματα) — structured notebooks described by Michel Foucault as "a material memory of things read, heard, or thought, offered as an accumulated treasure for subsequent rereading and meditation." The Session Brief is a hypomnema for the age of AI.


License

This work is licensed under Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0).

You are free to share and adapt this material for non-commercial purposes, with appropriate credit. For commercial use, contact the author.


Suggested GitHub Topics

ai-context · session-management · ai-methodology · context-engineering · ai-memory · multi-session · ai-workflow · notion · knowledge-management · llm-tools

Releases

No releases published

Packages

 
 
 

Contributors