I’m exploring options to do dictation for several tasks. Privacy is an issue and while I want the power of an AI to make the transcription actually accurate for some of these things there is no way I want to send the data to any offsite AI for review/analysis and transcription.
I’ve tried the Apple tools and they don’t cut it at all. Most of the stuff that I’ve seen recommended uses a cloud AI system.
The only one that seems to work offline or locally is SuperWhisper but I can’t seem to find a really in depth review of features. I thought MacSparky had done a series that included discussing SuperWhisper but I can’t seem to find it or maybe I just imagined it.
Can anyone point me to the MacSparky link and if anyone is using SuperWhisper as their primary dictation/transcription tool I’d love some tips and hints.
There’s a two-part Superwhisper for Mac series on ScreenCastsOnline. Part 1 was published on May 1, 2026 and Part 2 a week later. I haven’t gotten around to watching it yet, but I am a big fan of @Leegarrett’s work, and I’m sure it will be well worth a watch.
I think most of the dictation apps allow you to bring your own API key. For those that do, I imagine plenty of them also support doing it with locally hosted on device AI models, e.g. GPT OSS 20b. It’s only the cleanup step that you’d have to install your own models for. The transcription is already enabled by local models out of the box for most of these apps.
I"m using Superwhisper and very much enjoying at the moment. I’ve never been into dictation tools, but this one is really working for me. That said, lots of people using Wispr Flow as well and when I was trying both out, that was pretty impressive too.
You can indeed use your own API keys for Superwhisper and setup custom modes, which I have, so that you can really tailor it how you need. They have a generous free trial as well to play with.
If you want local models, they have those that you can download and the performance speed and accuracy I’ve found to be very impressive.
Here’s a comparison by another AI dictation service, Willow, which when you get to the end claims to meet the privacy feature of Superwhisper without the complexity but does send things to the cloud - described as privacy first cloud with no data storage.
It does confirm Wispr Flow does not have a local version.
Because MacSparky keeps praising Wispr Flow, I assumed that was the best dictation app for me. However, I prefer SuperWhisper. Here is why.
SuperWhisper runs its AI models on-device. Audio stays on your Mac unless you deliberately enable one of the optional cloud post-processing modes, which I leave off. This protects my privacy. SuperWhisper also runs without an internet connection. Wispr Flow requires one.
SuperWhisper does not rewrite what I dictate. It transcribes what I said, cleans up filler words and obvious stumbles, and handles punctuation, but it leaves my wording alone. As I have shared previously, I want AI as an editor, not a ghostwriter. Several of the cloud-based tools, including Wispr Flow, rewrite your words. I do not want that.
SuperWhisper offers a lifetime license at $249.99, with monthly and annual options as well. One payment, no subscription.
If you have no need for post-transcription-AI processing, why would you pay so much when you could just install the OpenAI Whisper model, locally? Something along the lines of what I outlined here. Feels like there a lot of “emperor’s new clothes,” type apps in this sphere, especially if just straight AI transcription is all you are after.
That is a fair question. I have not considered running a local model, and frankly, I lack the technical knowledge (I have no desire to deal with things like Homebrew and scripts) nor the inclination to learn. Also, SuperWhisper offers several models to choose from, and I can customize a model, which I have done. For example, I have created an “editor” customization. SuperWhisper works well for my purposes, and with the lifetime purchase, I am set for the foreseeable future.
I am using Voibe. It seems to work fairly well. I like the fact that it’s local only and the developer has said that an update is going to bring text displaying as you speak versus waiting until you’re done.
Well, I’m an enthusiast. I’ve never claimed to be a “power user.” In fact, I often admit that I’m among the least technically knowledgeable on this forum.
I am having a look at Voicelink for on-device transcription.
It seems like one needs to use AI enhancement to convert it into dictation - punctuation words like “comma” and “open parenthesis” etc. work as punctuation and not get transcribed.
The developer recommends using Groq and get-oss-120b for AI enhancement and this thread includes a very comprehensive prompt by ebastard.
I am a bit confused, however. If I am using AI enhancement in this way, doesn’t that mean the words are being sent off device to a server?
Using Superwhisper. I got the lifetime plan and it’s totally worth it’s. $250 ($149 for student)
It just works, and it’s accurate. Basically it does it’s job well.
With one-time payment, you don’t have to worry about checking API usage and hooking up an AI provider. So you can process large text etc without worrying about the cost. I process decently large files all the time.
It offers pretty good models out of the box. GPT 5.3, 5.2, Claude Sonnet 4.6, and a few others.
It is automation friendly with deeplink support and also a utility called macrowhisper which 5llows a tons of other stuff if you want to delve into it.
IMO it has the best UI/UX and offers real time transcription via parakeet.
iOS app is pretty good with support for action button triggers.
I think they are expanding rapidly, and they also offer iOS and Windows clients. Plus, if we have different operating systems and devices, it is totally worth it to get Superwhisper. You can also share it with your spouse, kids, and siblings based on a one time payment of $250, which is totally awesome. WisprFlow does not allow as far as I remember.
Raycast 2 has just been released (as a very clean beta) and it has dictation built in. I think it requires a paid account once it comes out of beta. Well worth it.
I have an annual subscription to Superwhisper and it’s a great investment. I mostly use the downloaded voice models and they’re good.
Also, this type of software is changing so much it’s neat that we can subscribe and switch. They all seem to work differently but are essentially the same.
I haven’t tried their beta yet. Looking forward for their implementation. If their processing is sent to cloud then it’s a bummer. At least with SuperWhisper I can have it all local.
Well, I’m an enthusiast. I’ve never claimed to be a “power user.” In fact, I often admit that I’m among the least technically knowledgeable on this forum. 