What is DOSI?

DOSI (Development Operational Session Instructions) is a governance protocol for AI-assisted software development. It defines how a human supervisor and an AI coding partner collaborate across structured development sessions.

The Problem

When you use an AI coding tool (Claude Code, Cursor, Copilot), every session starts cold. The AI doesn't know:

• What you built yesterday
• What architectural decisions you made last week
• Why you chose SQLite over Postgres
• What's blocked and what's next

You end up repeating context, re-explaining decisions, and losing momentum between sessions.

The Solution

DOSI provides three things:

1. A Session Protocol

Every session follows a lifecycle:

Boot    → Read state, read history, confirm objectives
Execute → Work, log continuously, capture ideas
Close   → Finalize docs, commit, embed into memory, snapshot

The AI reads the protocol on boot and knows exactly how to operate.

2. Operational Memory (OCD)

Every session produces an OCD (Operational Context Directory) — a self-contained context unit. The session document, plus every artifact created during the session — plans, schematics, diagrams, data, any file type.

Session 47 can read what session 1 decided. No context is lost.

3. A Knowledge Network (v0.3)

OCD files get chunked and embedded into a vector database. On boot, the AI queries the knowledge network with the session's goal and surfaces relevant history automatically.

"Last time you worked on the adapter system, you decided X in OCD-038 and refined it in OCD-051."

DOSI vs OCD

	DOSI	OCD
Is	The system (protocol + pipeline)	A context unit (directory)
Contains	How to operate	Everything from one session
File types	Protocol document	Any — .md, .svg, .json, .sql, diagrams, schematics
Updates	When the system evolves	Every session produces a new one
Goal	Structure the workflow	Feed the vector DB

Who Made This

DOSI was created by Bepely (GitHub) and battle-tested across 60+ development sessions building production software with Claude. It evolved through three major versions as the development practice matured.

Where Does It Run?

Anywhere. DOSI doesn't require a special repo or an HQ setup. Drop dosi/ into any project and start working.

The dosi/ folder is development infrastructure, not product code. You can:

• Gitignore it — keep it local, point a vector DB at it on disk
• Track it in git — version your sessions alongside your code (good for solo projects)
• Put it in a separate repo — clean separation between product and process

You don't need HQ. You don't need multi-project orchestration. One project, one dosi/ folder, one AI partner — that's the simplest DOSI setup, and it works.

HQ is just a pattern for when you scale to multiple projects. See HQ Pattern.

Core Principles

1. Separation of concerns — dosi/ operates ON the product, not IN it
2. Capture first, structure later — raw ideas in immediately
3. Autonomous with oversight — AI works independently, human steers
4. State is explicit — read on boot, write on close
5. Portable — works on any project, any AI tool
6. Cumulative — every session compounds the knowledge network