Typeahead - AI autocomplete for every app on your Mac
by•
Typeahead is the writing assistant that writes with you, not for you. It works in every text field on your Mac, helping you type faster and smarter with inline suggestions that appear as you write. It runs on a local AI model, works offline and keeps your writing on your device.
Replies
Typeahead
Hey Product Hunt 👋
I’m Sam, co-founder of Typeahead.
We built it because most AI writing tools still take you away from the place writing actually happens. You open a chatbot, paste text into a box, get something back, then spend time trying to make it sound like you.
That never felt like the right interface for writing.
Typeahead works inline in every text field on your Mac. As you type, it instantly suggests words and sentences right at the cursor, so you can move faster and think less about the tool. It feels more like your computer helping you write than using a separate AI product.
A big part of the product for us was making this local. Typeahead runs on a local AI model, works offline and keeps your writing on your device. We also wanted it to feel like software you own, which is why it's a one-time purchase rather than another subscription.
A few things that make it different:
inline suggestions, right where you already type
works across your Mac instead of inside one app
built to write with you, not for you
runs locally, your writing stays on your device
one-time purchase, free updates for life
Would love to hear what you think, especially from people who have tried chat-style AI for writing and found it broke their flow.
I'll be around all day replying to comments.
@samasante Honest question: inline suggestions you accept and keep typing is basically Copilot 2021, and code has since run off to full agents. Is "you still type" a deliberate bet — that writing is different from code because the hard part is voice, not output — or just the starting point before something more autonomous?
Typeahead
@haotian_wang5Â I was hoping someone would ask this! Definitely a deliberate bet, not a stepping stone.
With code, the output is the thing. It compiles or it does not, tests pass or they fail, so handing more of it to an agent makes sense because you can verify the result.
Writing is different. The output is the voice. The hard part was never generating words, it is making them sound like you and say what you actually mean. The moment you hand that to an agent you get something back that you have to read, reshape and re-voice, which is the exact loop we were trying to kill.
So you staying in flow and typing is not a limitation, it is the point. Tab when we get it right, keep going when we do not, and it stays your writing the whole way through. For writing we think augmentation beats automation, because the value lives in you staying the author.
Headless Dropshipping Starter
@samasante mate, this is awesome! It's really fast too which makes it feel really part of whatever app I'm using. Way to go!
The interesting edge case with system-wide autocomplete is how it handles context switching. does it know you're writing a Slack message versus a legal doc in Word and actually shift register accordingly, or is it one tone fits all? Also curious whether it reads the surrounding text in a field or just the last few words, because that gap is usually what makes suggestions feel off.
Typeahead
@fberrez1Â Great question. Context handling is something we spent a lot of time on. Rather than just grabbing the last few words, it reads the full surrounding text in the field, and we have built custom detection tuned for the apps people use most like Slack so it captures the right context for each one. That is what stops suggestions from feeling off.
It picks up the tone you are already writing in, so a quick Slack reply and a formal doc come out differently. And all of it runs on device, no screen capturing and nothing sent to the cloud. You can also set your own instructions to steer tone, with per-app prompts on the way.
Really impressed by this. Congrats on the launch. I’m kind of amazed at how fast the suggestions show up. They feel instant, and more importantly they’re actually useful. A lot of AI writing tools are slow or get in the way, but this genuinely helps me type faster.
Typeahead
@jonnotie Thanks Jonno, this means a lot! Speed is something we put a lot of work into. A suggestion that shows up even a beat late breaks your flow instead of helping it, so getting it to feel instant was a big focus. Hearing it actually helps you type faster is the whole reason we built Typeahead 😃
Really like the "writes with you, not for you" angle — that's the line I
keep coming back to when building my own AI writing tool. Most autocomplete
products try to take over and end up feeling pushy.
Two genuine questions :
1. How did you balance suggestion frequency vs friction? My biggest worry
with always-on autocomplete is that it interrupts flow when you actually
know what you want to type. Did you implement a "quiet mode" threshold or
is it always live?
2. The on-device model is a strong sell for privacy — what model size are
you running, and does performance scale with M-series Macs?
Bookmarked. Looks like one of the cleanest implementations I've seen in
this space.
Typeahead
@fabiendsh Thanks Fabien, really appreciate that.
It is always live but designed to feel invisible when you're not using it. Suggestions are just light ghost text, so when you already know what you want you keep typing and ignore it, Tab is the only thing that commits. When you don't need to type faster, you can pause for 30 minutes, an hour or until tomorrow, or switch it off in specific apps entirely.
The default is Gemma 3 4B running locally with Metal. You can drop to a lighter 1.7B model if you want more speed, or step up to Gemma 4 for more capability. So yes, it scales with Apple Silicon, the newer the chip the snappier it feels.
Hope you give it a try, would love to hear what you think.
PhotoFox AI
Interesting idea. But I would like to try it before purchasing it.
Typeahead
@nasimuddin01Â Check out our launch video for a few examples of how Typeahead feels in action! And be sure to follow us on social media for updates and more product news
Shadow
Nice launch. What got me is that it runs locally and works in every text field instead of pulling me into a separate chat window just to rewrite a sentence. The writes with you not for you angle feels right. Congrats Sam, followed you on X too.
Typeahead
@JayTheSong Appreciate it Jay. Keeping it local was a big focus for us. Your writing should not have to take a round trip through someone's cloud just to finish a sentence. Thanks for following along, watching Shadow too.
M1 air 8gb. chrome and cursor already hurt. does the model sit in ram all day or only spin up when i'm typing?
Typeahead
@novamaker01Â Typeahead is optimized to be extremely memory efficient. I am running the heavier Gemma 4 model right now and RAM usage is sitting at around 300MB, often less than a single Chrome tab, so it should work well even on an 8GB machine.
It loads once and stays ready in the background so suggestions are instant and quitting frees it straight away. Your Air should handle it well, and if our built in options ever feel too heavy you can load your own custom model. The models are getting more efficient by the day, so performance will also continue to improve over time!
An interesting app. I would like to see something like this for Windows.
Typeahead
@sergei_popovichev Thanks Sergei! Running locally and optimized for Apple Silicon is a big part of what makes Typeahead so fast, so Mac was the natural place to start. We'd love to release a Windows version though, so hearing the demand helps make the case.
The fact it's running a local model is a huge upsell for me. Excited to see how the product evolves - It works great, easy af to install. I do lots of dictation these days, but I find a totally different thought pattern when typing - this is a cool addition to the stack.
Typeahead
@hunter_hastings Thanks Hunter, glad it's working well for you! And totally agree, typing and dictation put you in completely different headspaces. Good to hear it's earning a spot in the stack.