Built a project to Automate my Job

I kept doing the same loop every morning: read the mail, work out what was actually work, then go and do it somewhere else. The deciding was the job, and nothing helped with it.

So I built Taskuary. It runs on my own machine and pulls everything inbound onto one timeline — Outlook, Teams, Slack, GitHub issues, scheduled SQL and cloud reports. An AI reads each message and says which of three things it is: real work, a question that only needs an answer, or noise. Questions come back as a draft in my voice, waiting for me to approve. Real coding work goes to the CLI agent I already use — Claude Code, Codex, bring your own — in a session I can watch and interrupt. Nothing sends or ships without me pressing the button.

The part I didn't expect to matter most: it learns. When I say "this isn't our task", it writes down why, and the next message like it never becomes a task again.

Local-first — SQLite on my laptop, my own API keys, nothing goes to a server I don't own. MIT licensed.

It's early — v0.2.1 — and in daily use on my own inbox. The edges are still being knocked off and breaking changes are possible before 1.0.

pip install taskuary → https://github.com/ldbumble/taskuary

I'd genuinely like to know: if you tried it, what was the first thing that confused you? That's the bit I can't see from inside it.