cenno

cenno is a macOS app that lets AI agents ask you questions through minimal floating panels — and get your answer back as structured data.

If you run MCP-capable agents (Claude Code, Claude Desktop, OpenClaw, Hermes, or any other agent that speaks the Model Context Protocol), cenno gives them a polite way to reach you: a small panel slides over whatever you're doing — without stealing your keyboard focus — you tap or type, and the agent receives {answer, via, elapsed_s} as the tool result. Every exchange is recorded locally, so your answers become your own queryable dataset.

Built for time-sensitive, low-friction interactions: mood check-ins, EMA questionnaires, quick decisions, reminders.

Fare un cenno — Italian: to give a small sign. To nod (mi ha fatto un cenno), to mention briefly (ha fatto un cenno al problema), to beckon silently (mi ha fatto cenno di entrare). That's the whole interaction model: agents make a small sign, you answer with one.

---

Install

Download the latest DMG (signed & notarized, Apple Silicon, macOS 12+), drag cenno to Applications. Or via Homebrew:

brew install --cask glebis/tap/cenno

On first launch cenno lives in your menu bar, registers itself to launch at login (you can turn that off in the tray menu), and waits for prompts.

Requirements: Rust stable, Node 20+, Xcode Command Line Tools.

npm install
npx tauri build --no-bundle          # unsigned dev binary → src-tauri/target/release/cenno

# Signed, installable .app + .dmg (your own Developer ID):
APPLE_SIGNING_IDENTITY="Developer ID Application: Your Name (TEAMID)" npx tauri build

Note for contributors: a plain cargo build binary loads the Vite dev server (port 1430) and shows a blank page unless npm run dev is running. CSP is only enforced in bundled builds — test security changes against npx tauri build output, never tauri dev.

Quickstart (60 seconds)

1. Install cenno (or download the DMG):

brew install --cask glebis/tap/cenno

2. Add it to any project's .mcp.json:

{
  "mcpServers": {
    "cenno": {
      "command": "/Applications/cenno.app/Contents/MacOS/cenno",
      "args": ["--mcp-stdio"]
    }
  }
}

3. Your agent now has an ask_user tool. Or try it from the shell:

/Applications/cenno.app/Contents/MacOS/cenno ask "Ship it?" --timeout 30

A panel appears; your answer prints as JSON. cenno auto-launches if it isn't running.

Agent skill

Want your agent to ask well — pick the right input kind, sensible flows, graceful timeouts — and to wire cenno into a project itself? Install the ready-made cenno skill. It bundles ask_user usage, custom 1–N scales, etiquette, and a setup mode.

Install via npx skills:

npx skills add https://github.com/glebis/cenno/tree/main/skills/cenno

Or just ask your agent (install via prompt):

Install the cenno agent skill from https://github.com/glebis/cenno/tree/main/skills/cenno — copy it into my Claude skills directory.

Once installed it activates on its own whenever a task calls for asking the user, running a check-in, or setting cenno up.

Why not just a terminal prompt or a dialog box?

• Focus-preserving: the panel is a non-activating macOS panel — it floats above your work but never grabs your keyboard. You answer when you glance at it, not when it interrupts you.
• Structured answers: agents get typed results (text / choice / scale / confirm), not parsed strings.
• Rich when it needs to be: tables, radio lists, a per-item rating matrix, and image pickers — rendered in the panel, not dumped as text into a terminal.
• Respectful by policy: pause it from the tray (15 min → until tomorrow), and it stays quiet automatically while the screen it lives on is in fullscreen. Suppressed prompts queue and replay; agents just see their normal timeout.
• Local, inspectable storage: every outcome lands in a local SQLite file (0600 permissions) and exports as JSON/CSV.

---

The ask_user tool

Param	Type	Notes
title	string	the question
body_md	string	optional markdown body
input.kind	text · voice_text · choice · scale · confirm · none	answer control
choices	string[]	for choice
flow	mood · question · ema · reminder · ambient	panel color theme
progress	{step, total}	dot pagination for multi-step flows
timeout_s	number	default 120
urgency	low · normal · high	how insistently the panel asks (and whether it speaks)
say	string	short line spoken aloud instead of reading the panel
muted	boolean	open silently regardless of urgency gating
device_hint	phone · ipad · watch	a proposal — your ~/.cenno routing policy decides
a2ui	array	optional rich layout (A2UI v0.9 messages, validated at the boundary; see docs/design/CONTROLS.md)

Returns {answer, via, elapsed_s, muted}, or {answered: false, prompt_id} on timeout. muted is the panel's final voice state and may be absent on older frontends.

Rich panels (a2ui)

For anything a title + a row of buttons cannot express, an agent can send an a2ui payload (A2UI v0.9, catalog cenno:catalog/v1) and get a composed panel instead. Beyond the standard text/row/column/button/field controls, cenno adds five:

以上为 README 的开头部分。其余内容请阅读原文。 查看完整 README ↗