Cotypist is fantastic because it feels simple, fast, and almost built into the operating system. I use Apple Voice Control for a lot of my typing, and Cotypist helps make that experience smoother and more efficient without getting in the way. It is one of those tools that quietly solves a real problem and makes everyday writing feel easier.
Honestly, I can’t think of anything right now. Any time I’ve had feedback or thought of something that could be improved, the Cotypist team has either fixed it or made an update that addressed it. That’s one of the things I really appreciate about the product — it keeps getting better.
I chose Cotypist over TypeAhead because it felt more seamless. It fits better into my workflow, especially with how much I use Apple Voice Control. I wanted something that helped speed up typing without adding extra friction, and Cotypist felt cleaner, smarter, and easier to trust day to day.
Cotypist is a true gem that makes typing text on my Mac dramatically more efficient and accurate. I appreciate that all of its magic is performed locally and, having used Timing for years, trust the developer behind this great app.
Cotypist is unparalleled. While there are other apps that can help with typing, I haven't seen any that are completely local and match the breadth of features that Cotypist offers.
I've been using Cotypist since early in the beta and it's been great. Daniel (the dev) has been very responsive to feedback and I get the feeling the product is well engineered. (I was not in any way incentivised to write this).
Hey everyone, I'm Daniel, the developer behind Cotypist.
First, a quick thank-you to the Product Hunt team. After Cotypist launched back in May, they reached out and invited me back for a featured relaunch. I'm honestly a little stunned by that, and very grateful to be here again.
A few years ago, I noticed I'd developed a weird habit: copying conversations into Visual Studio Code, just to get GitHub Copilot's inline completions, then pasting them back into the app I should have been writing in. After enough of that, it clicked: autocomplete shouldn't live in one editor. It should work wherever you write.
So I built Cotypist. It's smart autocomplete that runs locally on your Mac (no cloud, no API calls), in basically every app you type into. Install it, give it a minute, and you're writing faster everywhere on your Mac. No long setup. Tab to accept a suggestion, keep going. Words still sound like you.
You can download Cotypist today from https://cotypist.app; there's a free 30-day trial with all the features, and there's also a free plan for casual use after that.
During early access, Cotypist has become a daily driver for founders, marketers, support folks, novelists, physicians, academics, and long-time Mac users. People who type a lot of email, Slack, and AI prompts. Plus a long tail I didn't see coming: non-native English speakers, one-handed typists, and (this still blows my mind!) not one but two Neuralink brain-implant wearers.
What still surprises me about Cotypist, even after building it, is how often it feels like it's reading your mind. Or almost like a colleague finishing your sentences.
Happy to take questions about the product, where it works (and where it doesn't), what's coming next, or anything else. I'll be here all day.
—Daniel
@daniel_a_a one of the best desktop apps i tried for a while. i wish it was embedded in ios. but apple will never allow a keyboard that freedom. we already know that from dictation apps
@ourielohayon Thank you for the friendly words! Yes, iOS keyboard extensions unfortunately are very limited in what they can do. Here's hoping that someday this will be possible!
@daniel_a_a Many congratulations Daniel on the launch! :)
Daniel reached out to me, and I found Cotypist incredibly novel. We already have tools like Text Blaze for snippets, Wispr Flow and Aqua Voice for dictation, but this is different: it suggests what to write next while you’re typing, anywhere on your Mac.
The product clearly had real interest behind it since Product Hunt invited Daniel to relaunch, I was happy to support it fully through my hunt.
A few things stood out to me:
It runs locally. No cloud, no API calls, and it works offline on your Mac.
It’s fast. Predictions appear in real time, often with no noticeable delay.
It has a low-risk trial. There’s a free 30-day Pro trial, plus a free plan afterward.
It still sounds like you. Cotypist learns your voice, so the output feels like co-typing rather than AI writing.
It works everywhere. It’s not limited to one editor; it works across Mac apps like Mail, Slack, Notes, docs, and AI prompts.
It works when dictation doesn’t. It’s useful in places where speaking out loud isn’t practical, like libraries, meetings, or flights.
It’s built by someone deeply focused on the problem. Daniel has spent two years refining it, and that shows in the quality of the product and the support behind it.
Overall, Cotypist crosses two important thresholds at once: the suggestions are good enough that you actually want to accept them, and fast enough that they never interrupt your flow.
Give it a try and share your thoughts in the comments. :)
@rohanrecommends Rohan, thank you so much for hunting Cotypist, and for the support! You’ve put your finger on exactly what makes Cotypist special. I’m glad to hear that that is resonating, and am excited to see the positive response from the community.
@daniel_a_a Kudos on the launch. Quick question: what’s one unexpected real-world use or workflow where Cotypist has surprised you by making a big difference?
@swati_paliwal Cotypist's use as an assistive tool has definitely surprised me. I've heard from two Neuralink users with ALS as well as a quadriplegic who are using Cotypist to massively speed up how they communicate with the world. It's heartwarming to see how Cotypist can meaningfully improve people's lives in this way.
@daniel_a_a I'm a huge fan of Cotypist! I use Apple Voice Control to type most of the time due to my disability (born without hands) so I would love to talk to you about some ideas I have for getting them to work together. In the meantime, I'll be over here hitting tab, saving time typing! Thank you for making such a thoughtful and incredible app. It's definitely an app I don't want to use a computer without. Keep up the amazing work.
@hudsonperalta Thank you for the kind words, I am really happy that Cotypist makes a difference in your life! And I have some ideas for additional accessibility features; talk to you soon!
@daniel_a_a Local AI autocomplete is the right move for privacy. How did you handle the latency piece — was keeping responses snappy the hardest technical challenge, or was it more about model selection?
I was in on the early release of Cotypist and saw it had great potential. It was then over-eager in the same way that the usual auto-correct is, causing a lot of backspacing. That is now totally gone with the tab completion. Start typing and it will provide a suggestion and if you like it, just press tab, but if not just keep typing. There's some tweaking when using apps with competing auto-corrects but that's not hard. If you've ever been in a relationship with someone where you get to the point of being able to ccomplete each other’s sentences, this will feel familiar. This goes into the day one new computer setup toolkit.
@technocrat Thank you for the endorsement, Richard! I appreciate your support throughout the early access period and am glad to hear that the improvements I made have made a difference for you. Acknowledged on the conflicts with e.g. the built-in macOS autocorrect; I’ve been thinking whether to offer disabling macOS' built-in autocorrect when installing Cotypist, but didn’t want to mess with the user’s system settings too much. I’ll keep iterating on it, though!
the "no cloud, no API calls" part is what makes this interesting. most AI writing tools send every keystroke to a server, which means your drafts, emails, and half-formed thoughts are all sitting in someone else's logs. running inference locally sidesteps that entirely. practical question though, how large is the model and how does it handle the tradeoff between suggestion quality and system resource usage? i'd want autocomplete that's fast enough to not interrupt my typing flow, but local models can get heavy on older machines.
@shubham4real Hi Shubham, you've put your finger on exactly why I built it this way. Sending every keystroke to a server means your drafts and half-formed thoughts live in someone else's logs, and running inference locally is what avoids that completely. On your practical question, which is the right one to ask:
Model size. The default is Google's Gemma 4 E2B, running entirely on your Mac. It's a few gigabytes on disk, but only about 1 GB of it needs to be in active memory while it's generating. Even though it's a multi-billion-parameter model, Cotypist doesn't keep all of its weights resident at once: part of them are streamed in from disk only as they're needed, so they never have to sit in GPU memory. That's a big reason it punches above its weight, you get close to the quality of a much larger model for the footprint of a small one.
The quality vs. resources tradeoff. Cotypist matches the model to your hardware rather than running one heavy model everywhere. Lighter and older Macs get the roughly 1 GB E2B by default; stronger Macs (the M-series Max and Ultra chips) can step up to the larger E4B, around 2.5 GB in memory, for better suggestions. The heavier models are only offered on machines that can actually run them well, so you won't accidentally bog down an older Mac, and it winds down and frees that memory when you're not actively typing.
Speed. This is the part I've spent the most time on, because autocomplete only earns its place if it clears two bars at once: good enough that you usually want to accept it, and fast enough that you're never waiting on it. Suggestions appear in real time, usually with no noticeable delay, and they keep updating as you write. People run it comfortably even on a base MacBook Air. The honest test for your particular machine is the free trial, so you can feel the speed on your own hardware before deciding anything.
"Autocomplete shouldn't live in one editor" is such a clean framing — the copy-into-VSCode-and-paste-back habit is painfully real. The local-only choice is what makes me trust it; I build voice/chat agents and privacy is usually the first objection. Question for you Daniel: how does "in your voice" stay accurate when it can't phone home — does it learn per-app (my Slack tone vs my email tone differ a lot), or is it one global style profile?
@david_marko Cotyist learns your voice locally; that does not contradict the local-only choice. The adaptation is a mix of both "global" and context-specific. In addition, you can manually provide custom instructions (either generally or per app or website) to tailor Cotypist’s suggestions even further in certain contexts.
How do you keep the first suggestion after an idle pause from eating a cold-read penalty? Do you pin a hot subset or just accept the occasional slow first token? Great work, you guys are on the right path!
@artstavenka1 The current context is often relevant only for a few minutes at a time, which is easy to handle. And in the other cases, the occasional slightly longer wait time is usually acceptable.
@daniel_a_a I'm curious why Cotypist doesn't the Apple Neural Engine at least as an option. It's much more efficient than inference on CPU/GPU, which would significantly improve concerns about battery life. It would also eliminate the obnoxious "chirping birds" sounds from coil whine as I type (M5 Max).
@coreyward Up until recently, the Neural Engine was not available for high-complexity tasks such as running LLMs, let alone quantized versions of them. It remains to be seen how much that has changed with projects like anellm and Apple’s new Core AI frameworks. However, I am still doubtful that the Neural Engine would provide a low enough latency to be useful for generating autocomplete suggestions in real time. In addition, the Neural Engine would also draw power in a modulated fashion, which could leave it susceptible to coil whine as well.
@daniel_a_a I think you're underestimating the Neural Engine on even M3 Max chips. From experience, it's shocking how capable it is. I've never experience coil whine when using it on my machine, even when stress testing it (I was curious how much wattage it caps at), but it's possible other machines are different.
Edit to link this: https://huggingface.co/google/gemma-4-E2B-it/discussions/14
We're running Gemma 4 E2B on iPhone Neural Engine via CoreML at
28 tok/s decode (99.78% of ops on ANE, verified via MLComputePlan).
MTP heads would let us implement speculative decoding and push
toward 40-50+ tok/s on-device.
You won't believe it, but recently I was looking for such a tool. I couldn't find something that would really work - just an autocomplete, without any other unnecessary functionality. And also that it would work quickly, and wouldn't eat up half of the computer's memory. It seems that this is what is needed. I will definitely try it and give my feedback.
I also have questions about this:
- What is the weight of the model that handles the autocomplete?
- Will this work in any input or text field? Or only in specific applications?
- What is the context for autocomplete? Is it the text that is previously entered into the text field, or does the application scan the entire page?
- After the trial period ends, if I like everything, can I buy the app permanently with a single payment? Is there or is there any plan to do so?
@oleg_tsizdyn Hi Oleg, thank you for the comment! This is the person I built Cotypist for: someone who wants autocomplete that just works, stays out of the way, and does this as efficiently as possible. Looking forward to your feedback. To your questions:
What model handles the autocomplete, and how heavy is it?
The default is Google's Gemma 4 E2B, running entirely on your Mac. The model file is a few gigabytes on disk, but only about 1 GB of it needs to be in active memory while it's generating. Even though it's a multi-billion-parameter model, Cotypist doesn't keep all of its weights resident at once: part of them are streamed in from disk only as they're needed, so they never have to sit in GPU memory. That's a big reason it punches above its weight, you get close to the quality of a much larger model for the footprint of a small one. On stronger Macs you get the larger E4B model (around 2.5 GB in memory) for even better suggestions.
Does it work in any field, or only specific apps?
It's system-wide. It works in most native macOS text fields and many web and Electron ones too: Mail, Slack, Notes, the major browsers, document editors, even AI prompt boxes. It isn't tied to a fixed list of apps. A handful of apps don't expose the information Cotypist needs to place a suggestion correctly, and those are listed on the compatibility page, but the large majority of where you write is covered. One notable exception is Ghostty; I submitted PRs to improve their support for Cotypist, but there hasn't been much progress on that front.
What context does it use?
Mainly what's visible on your screen as well as some information, plus some additional information. You can also provide custom instructions to tailor the suggestions to your writing style. All of that is processed on your Mac, and nothing you type is ever uploaded anywhere.
After the trial, can I buy it permanently with one payment?
Cotypist is subscription-based, and I don't offer a one-time purchase. The short reason is that the subscription is what keeps it in active development and support, faster completions, bug fixes, more app coverage, rather than a single payment that has to fund years of future work. That said, you're not forced to subscribe: after the 30-day Pro trial there's a capable free tier, so if it fits how you write, you can keep using it without paying anything.
Thanks again for giving it a try, and please do send your feedback once you've lived with it for a bit.
Typeahead
Thank you for all your feedback and support over the years Sam, I'm glad to have you along for the ride!