The user interface is clear and easy to navigate. While there are no additional features that are nice to have, all the features I needed for this specific usage are present.

I use SuperWhisper to dictate to AI, and I find the transcription much faster than typing. It's incredibly accurate and catches nuance well. With typical iOS dictation, I wouldn't capture the correct words, but since I can add my own library, it captures specialized terms easily. The initial setup was easy.

The basic loop is what sold me, and it has not changed since day one. I hold a hotkey, talk, release, and clean text lands wherever my cursor is. Not a raw transcript with filler words and missing punctuation, but a finished sentence I would have typed myself. There is no app window to open and no copy-paste step, because the output is pasted straight into whatever field I was working in, whether that is Gmail, Slack, Notion, or a plain text editor. The whole round trip on a short message takes a couple of seconds, and after a few weeks the hotkey becomes as automatic as reaching for the keyboard. Modes are the feature that separates it from every plain dictation tool I tried before it. A mode bundles a voice model, an optional language model for cleanup, a prompt, and formatting rules into a preset I can switch between. The stock ones cover messages, email, and notes, and they already change the register noticeably: the same rambling sentence comes out as a tight two-line Slack reply in Message mode and as a properly structured paragraph with a greeting in Email mode. What I use most is the per-app activation. I attached my message mode to WhatsApp and Telegram and my email mode to my mail client, so the right formatting fires automatically based on where I am dictating. I stopped thinking about which mode is active weeks ago, and that is exactly the point. Custom modes go deeper than I expected. I write my own prompt, pick which language model handles the cleanup, and save the whole thing as a reusable preset. A few of mine that earn their place: - a mode for client emails that keeps my sign-off and expands a few abbreviations I always use - a mode for internal notes that outputs plain bullet points with no pleasantries - a mode that takes my spoken Italian and returns polished written English, which I use daily - a rough-draft mode with no cleanup at all, for when I want my literal words Being able to pick the model per mode matters in practice. For quick messages I run a small fast model so there is no waiting, and for anything client-facing I use a larger one and accept the extra second. On Pro I can also plug in my own API keys, so heavier cleanup runs through my own account rather than a metered allowance. The translation workflow deserves its own mention because it changed how I work more than any single feature. I think faster in Italian than I write in English, and Superwhisper lets me speak the thought in my own language and get correct English at the cursor. What used to be a two-step process through a separate translation tool is now one breath. The accuracy on accented English is also better than anything else I have used, including the built-in macOS dictation, which used to mangle every product name and proper noun I said. Vocabulary handles the rest of the proper noun problem. I entered my clients' names, my company name, and the abbreviations native to my work once, and they come out spelled correctly every time since. It is a small feature that removes a disproportionate amount of post-editing, because names were exactly the words I always had to go back and fix. Local processing is the part I point to when someone asks about confidentiality. The offline models run entirely on my Mac, so audio for sensitive client material never leaves the machine. On Apple Silicon the local models are fast enough that I do not feel the difference from cloud on ordinary dictation. I keep one fully local mode specifically for anything under NDA, and knowing that the audio path never touches a server is what lets me use dictation for that category of work at all. Voice into coding agents turned out to be a real workflow rather than a demo. I drive Claude Code and Cursor by voice now, describing the change I want in two or three spoken sentences instead of typing a careful prompt. Spoken instructions come out longer and carry more context than what I would bother to type, and the agents do noticeably better with more context. It is the use case I did not buy the app for and now use every day. Super Mode is the newer addition I was skeptical about and now leave on. It reads the context of the screen I am dictating into and adjusts the output to match, so a reply dictated into an existing email thread picks up the tone of the conversation, and a comment dictated into a document follows the formatting around it. It gets things right often enough that I only override it for edge cases, and when it misreads a situation I switch to an explicit mode and move on. I would not want it as the only way to control output, but as a default that removes one decision per dictation, it earns its keep. Push to talk is my default trigger and it behaves exactly as it should: hold, speak, release, done. There is also a toggle style for longer dictations, plus configurable shortcuts for the functions I care about. File transcription covers the occasional voice memo or recorded call I need as text, and the meeting assistant records and produces usable notes when I need a record of a call for myself. Pricing is fair for what it does. The free tier is genuinely usable with the smaller models, Pro is in single digits per month, and the lifetime license is the option I went for, since a one-time payment for a tool I use across Mac and iPhone is easy math. One license covers all my devices, and the 30-day refund meant trying it properly carried no risk.