BreatheReply hasn't launched yet. When it arrives, getting started is three steps — install it, add your model, generate your first reply.
BreatheReply hasn’t shipped yet — there is no installer to download today. Join the waitlist and you’ll hear about it first. When it does ship, Windows may warn that the publisher is unknown — that’s expected, the installer isn’t code-signed yet. Click “More info” → “Run anyway”. After setup it runs quietly in the background.
Open Settings → API config. Paste your own API key, or point to a local model (Ollama, LM Studio — no key needed). Verify, and you’re ready. Keep a copy of your key somewhere safe — once saved, the app can’t show it again.
In any chat, select the message you want to answer. A green button appears beside your selection — hover it and choose Generate reply.
BreatheReply doesn't sell you a model. You bring your own key — the same way you'd bring your own account. Here is what that actually involves, including the annoying parts.
A long password that tells a model provider “this request is from me, put it on my bill.” It is yours, not ours: BreatheReply never sees it, stores it nowhere but this machine, and adds nothing to what your provider charges you.
From any model provider with an OpenAI-compatible API. Sign in, look for “API keys” in the account settings, create one, copy it. You don’t have to pick a particular brand — BreatheReply speaks the same protocol to all of them.
Providers bill per token, and it depends on the model you choose — a smaller model is the cheapest option for short replies. BreatheReply takes no cut in between. Run a local model instead and it costs you nothing, with no key at all.
① The full key is shown once — copy it before you close that dialog. ② Once saved here, the app can no longer display it, so keep your own copy. ③ A key does not travel between computers — on a new machine you paste it in again, and the exported settings file never includes it.
Point BreatheReply at a local model — Ollama or LM Studio. No account, no key, and zero tokens leave your machine.
A model that fits on a laptop is usually much smaller than a hosted one, and this app leans on the model for structured work — reading the stance, naming what will trip you up, proposing three genuinely different strategies. Small models handle easy messages well and start to blur on hard ones — and that goes for smaller cloud models too, not just local ones. So: a smaller model is fine for quick, low-stakes replies; for the message you would otherwise stare at for an hour, reach for a stronger one.
Select meaningful text in any app and the button appears. Works with drag, box-select, double-click, triple-click, and Ctrl+A. Appears only in whitelisted communication apps, so it stays out of your code editor.
Copy text anywhere and press your global shortcut. The main window opens with the text already loaded — no pasting needed. Works even outside the whitelist.
BreatheReply first reads the scene. If a reply needs a fact only you have, it asks one short question instead of guessing, then writes a ready-to-send reply in your voice — and you can steer tone or stance with a click.
One conversation, several replies in a row: turn on continuous mode and each reply builds on what came before, so you don’t re-explain the context every time. It ends whenever you say so — or automatically when you switch to another app.
When what’s hard isn’t the wording but the position, it can ask before writing: three plain-language options, each combining the key fact with a stance. Prefer to see a draft first? That’s the default — switch stance from the result panel.
Under each draft it can point out what is likely to hold you up — being hit emotionally, a timing signal in the message, or a fact only you have — with one practical line on how to get past it.
By default it reads each message and decides the scene itself — nothing to pick. If you only ever use it for one thing, pin that scene in Settings — work, customer service, social media, or personal chat — for sharper replies.
Drafts open in an editor you can change directly — reword, add an emoji, keep what’s already right. Names it doesn’t know become tappable placeholders: fill one in, and every one of its kind in the message is filled at once.
Park a message with 5 preset slots (15 min ~ 4 hours) or a custom time. A separate list window keeps everything sorted, and one calm reminder arrives when it’s time. The reminder notification shows the first 60 characters of that message; Windows keeps notification history for a while — you can turn it off or clear it in your Windows notification settings.
One click sends the reply back into the input field it came from — no manual copy-paste. (Some apps need the cursor placed in the field first.)
Text you can’t select or copy — take a screenshot. Two on-device OCR engines read the conversation and reply just the same.
Main window, reply-later list, settings, screenshot review, and Highlight & Reply each get their own window — open one without losing your place in another. Closing just hides them; nothing is lost.
Ctrl+Shift+B), the reply-later list (Ctrl+Shift+L) and settings (Ctrl+Shift+,) — all customizable.Windows 10 or later. macOS is on the roadmap.