@hexxt/umm
v0.3.0
Published
A faster way to ask. Direct, structured answers in your terminal.
Maintainers
Readme
umm 🤔
A faster way to ask. Stop opening a browser tab. Stop starting a whole chat.
Just type umm and your question — get a tight, structured answer right in your
terminal.
$ umm why is the sky blue
why the sky is blue
Sunlight is made of every colour mixed together, and when it hits the air the
blue part gets knocked sideways far more than the rest — so blue reaches your
eyes from every direction at once.
- The same effect turns sunsets red: near the horizon light travels through
much more air, and the blue is scattered away before it reaches you
- Violet scatters even more, but there's less of it and our eyes barely see it
→ umm why are sunsets red · umm why is the sea blueNo quotes. No ceremony. No "..." around your question.
What it does
- Just ask —
umm how do i undo a git commit. No quoting, no flags. - Fixes your spelling —
umm photosinthisis→ answers photosynthesis, and tells you what else you might've meant. - Knows the word you forgot —
umm the thing that turns dc into ac→ inverter. - Answers in a scannable shape — a headline, the answer, a few details, and follow-up questions you can run next. Never a wall of chat.
- Pipes cleanly —
umm x | pbcopygets plain markdown, no color codes.
📦 Install
With npm:
npm install -g @hexxt/umm # or just: npx @hexxt/umm <question>— or —
Standalone binary (macOS · Linux · Windows) — grab it from Releases, or:
brew install hexxt-git/tap/ummFirst run drops you into a 10-second setup wizard. Re-run it any time with
umm --config.
Usage
umm <anything you want to know>
umm --config # pick your agent, answer length, sources
umm --raw # skip the pretty formatting (also automatic when piped)
umm continue # reopen the last answer in a real agent session
umm continue --pick # ...or choose an older one from your historyNeed punctuation the shell would eat (?, ', |)? Quote it:
umm "what's the deal with sourdough?"
When an answer needs a follow-up, umm continue hands you the full agent —
resuming the very session that answered where the agent supports it, and
replaying the exchange into a fresh one where it doesn't. It reopens in the
directory you asked from, and drops umm's skill entirely: that session is the
plain agent, free to edit files and think as long as it likes.
Asks are logged to ~/.local/state/umm/history.jsonl (newest 500).
How it works
umm is really two things:
- A single skill (
SKILL.md) — the entire brain. It defines how a question becomes a clean, terminal-shaped answer. - A tiny CLI — it hands your question and that skill to whichever coding agent you've got (Claude Code, Cursor, Gemini, opencode, Codex…) and renders what comes back.
That's it. The CLI does no magic of its own — swap the agent, keep the behavior.
Use it as a skill (no CLI)
The brain is portable. Drop it into any agent that speaks the Agent Skills standard:
npx skills add hexxt-git/ummMade for people who'd rather not Google. 🤔
