The system implements system-wide text automation by operating as an omnipresent OS-level accessibility daemon on macOS. Instead of confining inline code completion and generative suggestions to isolated software packages like VS Code or standalone web interfaces, Caret captures local screen buffer text structures across all open client spaces, including Slack, Jira, native mail clients, and terminals. By monitoring real-time active field contexts—reading purely active text metadata rather than parsing display pixels—the background state engine processes your writing habits locally, allowing you to instantly trigger contextually accurate multi-word autocompletes anywhere on your system simply by hitting the Tab key.
Hey everyone! Dan here, cofounder of Caret 👋🏼
We kept noticing the same frustration on our team. Typing in Slack, filling out a Jira ticket, writing a commit message, replying to an email. Our brains already knew the rest of the sentence. Our fingers just hadn't caught up yet.
Autocomplete exists in code editors. Everywhere else on your Mac? Nothing. So we built Caret.
What it does: Caret sits invisibly in the background and finishes your thoughts, anywhere you type on your Mac. Press Tab and it completes your sentence. One keystroke. No copy-pasting into ChatGPT, no switching windows, no waiting for a chatbot to load.
How it works:
You grant one accessibility permission and Caret reads the context of whatever you're working on: the app, the text field, what's on screen
No screen recording, no screenshots. It reads text, not pixels.
No integrations, no plugins, no per-app setup. It just works everywhere.
What makes it different: Behind the scenes, Caret turns your sessions into chains of thought, building a second brain that learns how you write, what you're working on, and what you're likely to say next. Those memories are stored locally on your Mac. The longer you use it, the sharper it gets. It stops feeling like autocomplete and starts feeling like an extension of your own thinking.
We're a tiny team and this is a v1. We love feedback and ship fast. Thanks so much for checking us out! 🙏🏼
trycaret.com
@dschwartz18 the 'tab to complete' interaction is exactly what’s been missing. i’m so used to it in vscode that i find myself hitting tab in slack and being disappointed when nothing happens lol. definitely giving caret a spin today.
@chrismessina I love you and Raycast! 🥇
@chrismessina
This is such a real pain point -- the context-switching tax is invisible until you actually measure how much time you lose to it. How does Caret handle sensitive fields like passwords or internal finance docs? Curious how you think about the trust layer when it's reading across every app. Also does the 'second brain' piece work across machines or is it purely local?
Congrats on the launch, rooting for you 🙌
@andrasczeizel Hey Andras! Thanks for the comment and believing in us!
In general about privacy - the information is local (sharing feedback is optional) and sent to LLM providers (who commit to not train on the data). We are thinking about features to blacklist or pause Caret. Would love to hear your thoughts!
The second brain is local, your memory and connections become more elaborate as you keep using Caret on you desktop. We see it continuing to more sensors and devices to truly become a second brain.
@andrasczeizel @dschwartz18
Also check out the memory tab on the toolbar app! it became my personal crm at this point 😙
@dschwartz18 @ron_adin1 Haha okay that actually sold me, checking it out now :D
@dschwartz18 @andrasczeizel YESSSS
@dschwartz18 That makes sense on the privacy side, local storage + optional sharing is a solid default. And the multi-device vision is exciting, rooting for you to get there. Good luck with the launch! :))
@andrasczeizel Thanks and looking forward to hearing your feedback 😁
every editor plugin promises AI anywhere you type. the value gap is always timing - knowing when to suggest vs when to stay quiet. availability is the easy part.
@mykola_kondratiuk The difference with Caret is context.
Editor plugins only see what's in the editor - Caret sits at the OS level, so it sees everything you've been working on across apps.
That's what makes timing solvable at all. If you know enough about what the user is doing, you can make a much better call on when to suggest vs. stay quiet.
That's a genuine differentiator. Most of my writing context lives across 3-4 apps - the reference doc, the brief, the Slack thread. Editor plugins never see any of that. OS-level access is the only way to bridge it.
@mykola_kondratiuk Exactly 🤘 Waiting to hear your feedback on the product!
@mykola_kondratiuk Thanks for the support!
I hit Tab in Slack all the time out of habit and nothing happens... exactly the frustration you're solving. Curious how it handles the trust question when it reads across every app, that's the first thing users ask us at Filently too.
@sophiafyi Exactly at this point I reply to most of my slack messages and update the team on progress with tab.
Regarding the privacy aspect we take a local-first approach. All context processing stays on your machine, and we only send data to LLM providers who have strict zero-retention policies (meaning they don't train on your data).
Thank you for the support!!!
@ron_adin1 That's a solid approach! Local-first processing is probably the most credible answer to the trust question right now. We went a similar route: extracted text gets deleted post-processing, nothing stored, nothing used for training. Still, it takes a few touchpoints before users actually believe it. Curious if you've found anything that moves the needle faster on that?
I spend a lot of time writing across Notion, Slack, Gmail, and various other tools, and the mental tax of switching context constantly adds up. Caret just removes a chunk of that friction. Tab completes what I was already going to type, without me having to switch to a chat window or rephrase anything.
What surprised me most is how quickly it starts to sound like me rather than a generic autocomplete. The local memory aspect is a smart call, and you can see it getting sharper over time. The memory tab is genuinely useful as a lightweight personal CRM on its own.
One keystroke. No friction. Congrats on the launch, looking forward to seeing where the multi-device story goes.
Caret is my new ⌘V
@rooly_ exactly! It's a whole new world!
looking forward to getting more of your feedback ✌🏼
@rooly_ I actually configured it as ⌘V 🤘 (you can do it in Caret's settings)