The team's AI stack, installable in one command

Problem

An AI-assisted development environment that only runs on the machine of whoever built it is not a team tool, it is an anecdote. Local models, documentation indexing, MCP servers and validation tooling are five separate installations, each with its own prerequisites and its own way of failing halfway through.

Solution

I packaged the whole environment into a self-installing folder launched with a single command. Before touching anything it checks the prerequisites and, if one is missing, aborts saying which: an install that stops midway leaves the user worse off than never starting.

It then installs the local models on Ollama, indexes the documentation, registers the MCP servers and puts the validation tooling in place. When it finishes it runs a check that everything actually responds, because an installer claiming to be done is not the same thing as a working environment.

There are no credentials inside the package. The ones needed are requested at install time and stay on the installing machine, so the folder can be copied and shared without auditing what is written inside it.

Architecture

Why it is built this way

It checks prerequisites before writing anything. An install that stops halfway leaves the machine worse off than not starting.

It verifies end to end at the finish. Until every piece is shown to respond, the installation is not done.

There are no credentials inside. They are requested at install time and never stored in the distributed folder, so it can be copied without auditing what it contains.

It is all PowerShell because that is what already exists on the team's Windows machines. Requiring a runtime install in order to install everything else is a bad start.

Models run locally through Ollama: neither client code nor SDK documentation leaves the machine.

Result

The full AI-assisted development environment installs on a fresh machine from a single command, with prerequisites verified beforehand and operation checked afterwards.