org-memory-os

This repository is a blueprint for building a shared AI memory system that an entire organization can use together. Think of it as a plan you hand to a coding assistant, and it builds the system with you, stopping to ask questions at every point where your specific company setup matters. The core idea is letting hundreds of employees and their AI tools read and write to the same knowledge base, without leaking private information, corrupting the data, or losing track of conversations.

The blueprint addresses a hard problem that appears the moment more than one person is involved. A single person can have an AI agent that remembers last week's work. But sharing one memory across two hundred people breaks that simple model. The README lays out exactly what breaks: identity is no longer a simple yes or no, multiple writers need strict coordination, and a single poisoned entry can spread to everyone. The guide walks through these failures and their fixes.

The repo is not a framework to install. It is a single Markdown build guide with an agent protocol. The guide contains thirteen pillars, seventy decision forks, and a build order of sixteen milestones. Each milestone must be proven with a real command and its actual output before it counts as complete. A single team can run a pilot at milestone nine, while turning it on for the whole company requires milestone fourteen, and regulated industries like banks add a final milestone fifteen.

To use it, you open the build guide, confirm six preconditions including a data classification policy and legal sign off, then paste the file into your coding agent. If any precondition is missing, the guide tells the agent to stop. The README notes that the later milestones hold most of the real cost and risk. The full README is longer than what was shown.