The Wizard's Pick

A local-first terminal assistant for authorized security testing. It uses a local model by default, keeps the session in SQLite, and installs with pip install wizards-pick.

What it is

Pick keeps assessment context, the model conversation, command results, findings, and reports in a local SQLite workflow. By default it talks to a project-local Ollama server at 127.0.0.1:11435 and a DeepHat-V1-7B model. One loaded model handles planning and reporting, so there is no second model sitting in memory.

It proposes shell commands. I only point it at systems I own or have written permission to test. Pick does not add a scope matcher, denylist, or sandbox. The account, machine, and engagement are what keep a session inside its bounds.

Why I built it

I wanted a helper I could use on systems I am allowed to test without sending the session to a hosted model by default. Keeping the usual path on loopback made that possible, and it forced a plain design: parse the model's command proposals strictly, and fall back when a small model drops a command into a code fence instead of structured output.