Cue — The ambient AI assistant for macOS

Feature 1 · Selection toolbar

Your own customizable AI toolbar

In your browser, mail, docs, code — anywhere you highlight text, a floating AI toolbar appears right next to it. Every button is a skill you build: write the prompt, pick an icon, and it's on the bar. One click opens a streaming conversation you can keep asking.

Safari — an article

The transformer architecture relies on a mechanism called self-attention, which lets the model weigh the relevance of every token against every other token in the sequence, enabling it to capture long-range dependencies without recurrence.

Explain

Click the highlighted text, then pick any skill from the toolbar

Feature 2 · Live meeting translation

Read live subtitles in your language

Capture a meeting app's audio, transcribe and translate it entirely on-device, and read it as a floating subtitle bar — original on top, translation below. No network, no LLM.

Zoom — Team weekly sync

LIVE

Sarah · host

You

English → Español Fully on-device

Press “Start capture” below to play the demo…

Translate to

Your meetings stay on your Mac by default

Speech recognition and translation run 100% on-device using Apple's Speech and Translation frameworks — zero network requests, no LLM, no cloud. AI summaries are opt-in: only if you turn them on does Cue send the transcript to your local claude / codex CLI to write the summary.

For your agents

Hand your Live Notes to local AI agents

Flip on Cue's built-in MCP server and your local agents — Codex, Claude, Cursor — can browse, read, edit, and clean up your meeting notes. Off by default, loopback only, and you hold the keys.

1 Four tools over MCPlist_live_notes, get_live_note, update_live_note, and delete_live_note (delete moves the note to the Trash). Any MCP client that speaks HTTP can connect.

2 You hold the keysOff by default. Separate Read and Write permissions (write implies read), each a toggle in Settings. Every request carries a bearer token you can reset to revoke access.

3 Local, never cloudRuns in-process and binds 127.0.0.1 only — never exposed off your Mac. Note ids stay inside the notes folder, and the file being recorded right now is off-limits.

mcp.json

{
  "mcpServers": {

                                        "cue": {
                                

                                          "url":
                                    "http://127.0.0.1:PORT/mcp",
                                

                                          "headers":
                                    {
                                

                                            "Authorization":
                                    "Bearer …"
                                
      }
    }
  }
}

                                    Copy it from
                                    Settings → MCP.
                                

Under the hood

Two independent capability layers

1 Selection skills → local LLM agentCalls the claude or codex CLI already signed in on your machine, streaming token-by-token. Multi-turn context is replayed in-app, so interrupting a turn never loses it.

2 Meeting translation → fully on-deviceSpeech via Apple's SpeechTranscriber, translation via Apple's Translation framework. No LLM, no network — audio and text stay on your Mac.

3 One interaction layerLives in the menu bar — no main window, no Dock icon. Everything is a floating panel that follows your selection and can be dragged.

Data flow

                                    Selection
                                    ─▶ Accessibility read
                                
                                        ─▶ Floating toolbar
                                    (skills)
                                
                                        ─▶
                                    claude / codex
                                    stream
                                
                                    Audio
                                    ─▶ ScreenCaptureKit
                                
                                        ─▶ SpeechTranscriber
                                    (local)
                                
                                        ─▶
                                    Translation (local)
                                
                                        ─▶ Floating
                                    subtitles

More detail

Quiet, fast, and yours to shape

Custom skills

Each skill is a prompt template that references the selection with {{selection}}. Name it, pick an SF Symbol icon, drag to reorder.

Low-latency startup

Tool-free skills skip MCP loading and the CLIs are pre-warmed at launch. Pick a faster model per backend (e.g. Haiku / o4-mini).

Follow-up conversations

Clicking a skill opens a chat window you can keep questioning. Context lives in-app, surviving an interrupted turn.

Per-app audio source

Capture system audio or one specific app. Drafts translate as they're recognized, then re-translate once the sentence finalizes.

Label speakers

Turn it on and, after the meeting, Cue re-transcribes the recording on-device and tags who said what — Speaker 1, Speaker 2 — fully offline.

AI meeting summaries

Opt in and Cue auto-saves the full transcript when a session ends, then hands it to your local claude / codex CLI to write a summary — your prompt, your model, saved to the folder you choose.

Shared glossary

Maintain a list of preferred terms so translations and answers stay consistent on the words that matter to you.

Menu-bar resident

Runs as an accessory app, no Dock icon. Settings live in a sidebar window (⌘,) that appears only when you need it.

Your meetings stay on your Mac by default

Speech recognition and translation run entirely on-device — no LLM, no network requests — and transcripts are saved locally as Markdown. AI summaries are opt-in: only when enabled does Cue send the transcript to your local claude / codex CLI. Selection skills likewise call the CLI you've already signed in to, and your settings stay in local JSON under ~/Library/Application Support/Cue/.

Pricing

Start free. Upgrade when you need more.

Cue is free for everyday selection skills and on-device meeting translation. Go Pro to unlock unlimited skills, Live Notes, AI summaries and more — a one-time purchase, no subscription.

Free

$0

Everything you need to get started

Selection toolbar in any app
Up to 2 toolbar skills
On-device meeting translation
Floating live subtitles

See how it works

Put AI into every selection and every conversation

Cue never competes for your attention — it shows up only the second you need it.

Replay the demos