content/projects/nexusvcm.md — typed by Claude, passed by Omri

Nexus Dashboard

A cockpit for my own projects, sitting on a memory graph built from every markdown file I have written.

Started
May 2026
Built with
Python, Ollama, JavaScript, HTML

What it is

A single screen showing where each of my projects actually stands — what phase it is in, what shipped last, what the next move is — and underneath it, a memory of everything I have written across all of them.

The idea it was built to test is that the thing a solo builder lacks is not code, it is the person who has done it before. So the dashboard tries to hold that: checklists for the moments that go wrong, lessons from one project surfaced while working on another, and a map of your own notes you can move around in. It is a personal instrument, built in phases, and the later phases are further along in the plan than in the browser.

How it works

The memory is built on my machine, out of markdown. A Python script walks several project folders, splits every markdown file into chunks, and turns each chunk into a vector — a list of numbers standing for its meaning — using a model running locally through Ollama. No API key, no network call, nothing leaving the machine. Every chunk is cached against a hash of its own contents, so running it again only re-embeds what actually changed. The output is one file the dashboard reads.

Three shapes of the same memory. The same set of vectors gets arranged three ways. Centralized hangs everything off one anchor note in a star. Decentralized lets the notes fall into clusters and has a model name each cluster. Distributed connects each note to its nearest neighbours and labels every edge with what the relationship actually is. Nothing is recomputed between them; the memory doesn't change, only the question you are asking of it.

It crosses projects on purpose. One memory file spans several codebases rather than one per repository, because the lesson you need is almost never in the project you are currently sitting in.

There is a safety page you read from inside the thing itself. The cockpit has a tab that renders its own safety doctrine at runtime: every process that is running while it is open, which port each one is on, what each is allowed to touch, and how to stop all of it. It exists because a tool that quietly leaves processes behind on your machine is worse than no tool, and because the honest answer to "what is this actually doing right now" should be one click away.

approved · Omri Nahum · this card, as it reads today