Vinculum Docs

What is Vinculum?

Vinculum means bond or bar in Latin — the horizontal line that binds. It's the shared memory layer that AI sessions write to and read from, so parallel agents can coordinate without ferrying context through chat.

Every decision, every directive, every implementation gets written as a typed entry in the graph. Sessions are ephemeral. The graph is not. When a grunt finishes work and its window closes, the implementation entry remains — readable by the reviewing lieutenant, the colonel, and every subsequent session that touches that part of the project.

This is the substrate not scaffolding distinction: read more →

How it works

You connect Vinculum to your AI client via MCP. From there, your AI sessions get a set of tools: write, read_entry, claim_work, spawn_grunt, and dozens more. Those tools are how your AI reads from and writes to the graph.

Work flows through a five-role model — Colonel, Major, Lieutenant, Sergeant, and Private — that all answer to you, the General. Colonels and Majors translate your intent into routed work. Lieutenants decompose a brief into worker directives and commit the result. Sergeants and Privates execute those directives. The General is the human; everything below is an AI session.

Quick links

• Five-minute quickstart — fastest path to a working setup
• Substrate not scaffolding — why the graph IS the memory
• The five-role model — Colonel, Major, Lieutenant, Sergeant, Private (you are the General)
• Claude Code integration — one command to connect
• Conventions browser — live from the graph that documents itself