Wispr Flow is an AI iPhone keyboard that transcribes what you say

This is a bit of a gimmick, but it’s also kind of cool. I didn’t type a word of what you are about to read. I dictated everything in this text into Wispr Flow, a cool dictation keyboard I have been trying for a few weeks. Here’s how it did: warts and all, with no hand-made corrections.

When Whisper contacted me to pitch their Speech-to-Text AI keyboard, I was a bit skeptical. Even Apple has tried to deliver a native dictation punctuation feature, but it’s proven quite hit or miss.

It also didn’t help that at first, the workflow to actually use the keyboard seemed a little fiddly. But as it turns out, what they did to get around iOS limitations and offer the closest thing currently possible to an AI-based transcription keyboard is pretty clever.

How it works.

Wispr Flow works based on what they call Flow Sessions, which are windows of time you let the app access the microphone. In settings, you can configure the app to automatically end Flow Sessions immediately after 5 minutes, after 15 minutes, after 1 hour, or never.

Once you tap Start Flow from the app’s third-party keyboard, it takes you to the full-blown Wispr Flow app, activates the Flow Session, and then hops you back to where you were. From then on, you can tap the microphone button and it’ll start capturing what you’re saying. When you’re done, tap the check mark button and end the transcription session. You can also tap the menu button on the left to manually turn off the flow session.

Inside the app, you’ll find:

– Your entire dictation history

– A quick notes feature

– A tab to create a custom dictionary of specific words that you want the app to learn how to write and replace in the dictation sessions

In settings, you can also set your preferred languages, in case you’d like the app to work with different languages than the ones you set as default on your device

Worried about privacy? Me too. So I asked them.

Wispr Flow‘s intelligence is based on a mix between Meta’s Llama and OpenAI’s Whisper models. The company told me that as they expand their feature set, they may bring in other model providers. Still, they will always vet them carefully for trust, security, and compliance and only work with providers who agree to the same zero data retention standard. They also assured me that Whispr does not store or share personal typing data with OpenAI or any kind of third party for training purposes unless a user has explicitly opted in

I also asked them if any of the recent WWDC25 announcements could help reduce the fiddliness of using the app, and they said that they’re exploring ways to use new APIs to integrate Flow to features like shortcuts and action button. As for Apple’s newly announced local models, they’ve got their eye on it, but are sticking to the current model providers for now.

All in all, Wispr Flow seems like an app trying its best to offer good, reliable, and private AI-based transcription despite technical limitations on iOS. The accuracy really impressed me, and if I were a voice notes guy or depended on this for accessibility, I would be all over it. It also works on MacOS and Windows, and has a waiting list for Android, Linux, and the web. It has a free plan with 2000 words per week and Pro and Teams plans with 20% off for annual subscriptions. It also offers custom enterprise plans, and you can check out all the details on their website

Best AirPods deals on Amazon