org-memory-os

# org-memory-os

> A vendor-neutral blueprint for building one shared, permission-aware, auditable AI memory across an entire organization: a system where every employee's agent reads and writes the same institutional knowledge, over whatever surface they already work in, under one compaction and degradation-avoidance discipline. You paste the build guide into a coding agent (Claude Code, Codex, Cursor, or an SDK) and it builds the system with you, stopping to ask at every decision your cloud, identity provider, and regulator would answer differently.

org-memory-os is a documentation-plus-architecture project, not a framework to install. It is the multi-user sibling of RAG-OS (a blueprint for a personal always-on AI agent). Where RAG-OS builds one agent for one developer, org-memory-os keeps the parts of that design that survive an organization and replaces the parts that break: the single-writer database, the single trusted operator, the single conversation window, the single budget, and a threat model that assumed a trusted corpus. The founding rule is that durable state lives in files and a database whose authority is explicitly ordered (an append-only Postgres claim ledger is truth, git-Markdown is a projection of it, the hybrid vector-plus-keyword index is a rebuildable cache over the Markdown, and model transcripts are disposable), the model sessions are ephemeral, and the coordinating layer holds zero model context.

The guide is organized as thirteen pillars, seventy decision forks (each with a recommended default and its alternatives), thirty-eight anti-patterns with the design rule that prevents each (five carried from the personal system, thirty-three new at org scale), and a sixteen-milestone build order (M0 to M15) where every milestone ends in a definition of done proven by a real command and its real output. A defensible pilot for one team ends at Milestone 9; company-wide requires through Milestone 14; regulated banks and insurers add Milestone 15. The hardest and most expensive subsystems are identity and authorization done freshly, the legal retention-and-erasure machine, the curation staffing that keeps the corpus from rotting, and model-risk governance.

## Core documents
- [Build guide](https://raw.githubusercontent.com/csnyder256/org-memory-os/main/BUILD-GUIDE.md): the full architecture as thirteen pillars, cross-cutting principles, thirty-eight anti-patterns, and a sixteen-milestone build order, written to be handed to a coding agent that asks the humans at every decision fork.
- [README](https://raw.githubusercontent.com/csnyder256/org-memory-os/main/README.md): overview, the personal-to-organizational break table, comparison to enterprise memory and search products, and FAQ.
- [RAG-OS](https://github.com/csnyder256/RAG-OS): the personal-scale sibling blueprint this one derives from.

## Topics
- Organization-wide AI memory, shared agent memory, enterprise agent memory, permission-aware RAG, ACL-aware retrieval, multi-tenant agent platform, Model Context Protocol, ReBAC, Zanzibar, OpenFGA, SpiceDB, OIDC, SCIM, agent identity, on-behalf-of delegation, hybrid search, reciprocal rank fusion, pgvector, cross-encoder reranking, temporal knowledge graph, supersede-not-delete, bi-temporal memory, context compaction, degradation avoidance, prompt caching, prompt-injection defense, memory poisoning, agent security, Rule of Two, budget governor, LLM cost attribution, chargeback, GDPR right to erasure, crypto-shredding, legal hold, WORM audit, ISO 42001, EU AI Act, model risk management, knowledge management, institutional memory.

## Repository
- [GitHub repository](https://github.com/csnyder256/org-memory-os)