Cue

Suggests, proofreads, replies, remembers directly in Mac

Mac

Writing

Artificial Intelligence

Hey Product Hunt 👋 Cue is free for 7 days with everything switched on, no card and no account. For launch it is 50% off with the code CUE until 17 August, so $14.50, paid once. Here is the idea. You already type all day, in maybe six different apps. Cue sits in whichever one you are in right now, learns you as you go, and takes over as much of the writing as you feel like handing it. Nothing to open, nothing to paste. Let me walk you through it in the order you would actually meet it. ━━━ 🧠 First, it starts learning you ━━━ Cue reads the conversation you are in and quietly keeps what matters: your week, your plans, the places you go, the people you write to, and the way each of them talks to you. Say to a friend on WhatsApp that you have a movie tomorrow at eight. Later someone asks in Slack whether you are free tomorrow evening. The reply Cue offers already knows you are busy and suggests another night. You never told it twice. One rule keeps it safe: only your own facts travel. What Tom told you stays in Tom's conversation and never surfaces in Sara's. And you can see all of it. The memory is a browsable list you can edit, and delete one line at a time. ━━━ ✍️ Then it starts writing with you ━━━ Now that it knows something, it can help. Pause in the middle of a sentence and the rest of it appears in grey, right at your cursor, in that app's own font. Tab takes one word if you only half agree. Another key takes the whole thing. Escape and it is gone. Click into an empty box and there is already a full reply waiting, built from the conversation above it and what Cue knows about the person you are answering. That is the fun part: you stop writing replies and start approving them. ━━━ 🩹 Then it cleans up after you ━━━ Stop typing and Cue rereads what you wrote. Red underline means a real mistake. Blue means it is correct but clumsy. Point at either and you see your whole message with the change already made, so you are saying yes to a result rather than gambling on one. One tap fixes that one. Hold the key and a ring fills, and when it completes every fix lands at once, in a single edit that leaves your bullets, line breaks and spacing exactly where they were. Hold the same key when there is nothing underlined and the ring turns blue instead: that is you telling it to look again, right now. Same gesture, and it works out which one you meant. ━━━ ⚡ When you want something specific ━━━ Sometimes you know exactly what you want done. One shortcut opens a small panel next to your cursor: Rephrase, Polish, Formal, Casual, Shorten, Translate. Return runs the first, a number runs any other. Write your own next to them. Each gets its own instruction, its own model, and its own shortcut, so the thing you do twenty times a day is one key. If you keep retyping the same wording into your prompts, save it as a snippet and drop it into any action with {braces}. And when you already know what you want, a second shortcut skips the panel completely and writes straight into the field. Two gestures. That is the whole app. ━━━ 🎛️ A different model for each job ━━━ Cue does four separate things: suggestions, proofreading, memory and actions. They want opposite things from a model. Suggestions need to be instant. A reply can afford three seconds if it comes back better. So you give each job its own model. A small fast one to finish sentences, something stronger to write replies, whatever you like for the rest. Pick once and forget it, or tune all four. Where those models run is up to you too. Download one and Cue runs it inside itself with no account and no key, nothing leaving your Mac. Or bring your own Anthropic, OpenAI or Gemini key, or point it at Ollama or any OpenAI compatible endpoint. Keys sit in the Keychain, requests go straight from your machine to the provider, and there is no server of mine anywhere in the path. ━━━ 🧾 It never asks you to just trust it ━━━ Under every reply, Cue shows its receipts: how many messages it read, whose voice it matched, which remembered facts it used, and what it knew about your schedule. An eye icon opens the raw text it took from the app. You also watch it work. No spinner. The panel edge lights up and the words arrive one at a time, so you can bail with Escape the second you see it going wrong. Something that knows this much about you should be able to show you its working. ━━━ 🔒 And a lot of places it simply will not go ━━━ There is a blocklist of apps Cue is not allowed to read, and it ships with the obvious ones already on it. Add any app you like, remove any you do not mind. Beyond that: it will not open over a password field, it stays out of terminals, code editors and password managers by default, and it reads conversations through the macOS Accessibility API rather than by taking screenshots of your screen. The memory is a file in your own Library. Never uploaded, never synced to anything of mine, never used to train anything. No account, no analytics, no tracking. ━━━ Native, a few megabytes, universal for Apple silicon and Intel, Developer ID notarized, macOS 13 or later. It updates itself. I am here all day 🙂 If you try it and the memory gets something about you wrong, that is the feedback I want most.