I build open tools for understanding your data

Your agent re-learns your database every session and never learns what a colleague would tell it in thirty seconds — which column is the real one, which join surprises people, which field died in the migration. dbprint measures the half a database can report, gives you one place to write down the half it can't, and puts both in front of the agent.

A coding agent worth using needs your whole toolchain, and handing it that on your own machine is a bad idea. claudebox gives each session a disposable container with everything the agent needs and only as much of the host as you allow, behind a web interface that renders the conversation — code, diffs, diagrams — with panels for tasks, queued follow-ups, and forking at any point in the history.

A shell setup is a decade of pasted fragments in whatever order they landed, and a new machine means doing it again by hand. profile.d turns each piece into a plugin listed in one file, runs them through a staged lifecycle so order is explicit, and puts a new machine one install away.