Hi everyone,
Has anyone here tried an app called Nexy?
an AI-Powered Voice Assistant Empowering Computer Control for the Visually Impaired
it can:
Control a Mac using voice commands
Understand what’s on the screen
Tell you where buttons or elements are located
Open and manage apps hands-free
Help with messages (write, read, summarize)
Perform browser tasks, including filling out forms
Do real-time web searches
Here’s their site:
Youtube channel: https://www.youtube.com/@Nexy-p6b
Official Verification Record — Nexy Release (March 2, 2026)
This is my fb https://www.facebook.com/SergiyZasorin1
this is my linkedin: https://www.linkedin.com/in/sergiy-zasorin/
this is my email: zasorin.sergiy@gmail.com
To download the app:
https://github.com/Seregawpn/Nexy_production/releases/download/App/Nexy.pkg
Comments
API?
I'm one of those using AI mostly via API, so I got API keys both from Open AI, Gemini, Claude, and lately also from OpenRouter which is a suprisingly smooth platform for accessing all sorts of AI models from a vast collection of different providers. Would it be possible to use Nexy with an API key for connecting to an AI provider, e.g. Open AI, or is this something you could consider adding in the future?
Answer to Cliff
Hi Cliff,
That’s a great question.
At the moment, Nexy doesn’t support API keys, and this was a deliberate decision.
I’ve spent the last two years building Nexy and optimizing AI costs. The challenge is that a single voice command isn’t just one API call. It typically requires 4–5 LLM requests, each carrying a large amount of context (screenshots, UI state, conversation history, tool descriptions, etc.), often around 20–30K tokens. As a result, even a simple task can consume around 100K tokens.
For someone using Nexy actively—around 100 commands per day, which is a realistic usage pattern—that’s roughly 10 million tokens every day. With today’s high-quality reasoning models, API costs can quickly grow into hundreds of dollars per month, depending on the model and provider.
I didn’t want to build a product that would only be affordable for a small number of people. My goal is the opposite: I want people to simply use Nexy without worrying about tokens, API bills, or unexpected costs. I want you to spend your time getting things done on your Mac, not managing AI usage.
That’s why I chose to build Nexy around a ChatGPT account. If you decide it’s useful for you, you pay one predictable subscription to ChatGPT and simply use Nexy without thinking about anything else. Our goal is to keep Nexy itself affordable as well (around $5/month) so that as many people as possible can benefit from it.
That said, API support is something we’ll continue to evaluate in the future if we can provide it without significantly increasing costs or complexity for users.
Thanks again for the suggestion!
I wanted to share a quick update on Nexy.
Hi guys!
I wanted to share a quick update on Nexy.
We’ve been expanding what Nexy can do beyond everyday Mac control. You can now use your voice to work with tools like Google Sheets, Google Docs, Slack, and HubSpot.
The goal is simple: make everyday work on a Mac faster and easier for blind and low-vision users.
I’d love to hear what you think and what other work tools you’d like Nexy to support.
https://youtu.be/C5r369UR9Xc?si=jkS3C4mz4-Pz1WOB
One more update on Nexy.
Hi guys!
I wanted to share another small Nexy update and show what it can now do with Google Docs, Google Sheets, and Apple Mail.
Watch the new demo
I’d also really love your feedback. What would you like to see in the next videos? What problems or tasks would you like Nexy to help you solve?
Your feedback will help me decide what to build and demonstrate next.
https://youtu.be/ptc0crPprUU?si=PSYXFvPxol4_F37e
Nexy Update: Nexy works with emails
Hi guys,
I’d like to share a new Nexy update and a short demonstration of how blind and low-vision Mac users can manage email using voice commands.
In the video, Nexy can:
Read unread emails and identify important messages.
Find personal emails or a specific message without knowing the sender.
Read complete email conversations and threads.
Prepare, edit, send, and reply to emails.
Find someone’s email address using only partial information.
Read a verification code received by email and help the user sign in to an account independently.
Watch the demonstration here:
https://youtu.be/RuAgJCdSRB4?si=fgqk1mtk3W96sj_B
Join Nexy — Only 5 Free Testing Spots Left
Hi guys,
I wanted to share a quick Nexy update. Nexy is a voice assistant for Mac that helps blind and low-vision users control and automate their computer using voice.
We currently have only 5 free testing spots left, so if you’d like to join Nexy and try it for yourself, now is a good time.
Download Nexy:
https://github.com/Seregawpn/Nexy_production/releases/download/App/Nexy.pkg
Installation guide:
https://youtu.be/x18Xxz_sacQ
If you’d like to see how our users are using Nexy, how we use the product, and what Nexy can do in real situations, you can find all of our demos and detailed videos here:
https://www.youtube.com/@Nexy-p6b
If you’d prefer more personal help or have specific questions, you can email me directly at sergiy@nexyai.app.
If WhatsApp would be easier, just mention it in your email and I’ll be happy to share my WhatsApp contact as well.
question
since this app uses AI moddles, can apple intelagence foundation moddles be added? nowadays, all apple silicon macs have AI, and it's ability depends on what chipset you have. . now, my mac is an forteen inch M 5 MacBook Pro, and that particular SOC has nural excellarators in each core. one would think that would be powerful enough to run Nexi. and for those luckey enough to have forteen and sixteen inch macbook pros with the higher end m 5 pro, and m 5 max chipsets, nexi could do even better on those.
Answer to Daniel
Hi Daniel
Good question.
I haven’t looked deeply enough into Apple’s Foundation Models yet to give you a confident answer, but I’m going to look into it properly.
My first thought is that some simpler requests could probably be handled locally. The bigger question is whether Apple’s model can work well with the tools Nexy needs, like MCP integrations, other plugins, and Computer Use.
If it can understand those tools, choose the right actions, and work reliably with things like the DOM and Accessibility AX, then it could be very useful. If not, we would probably still need an external cloud model for the more complex tasks.
So I don’t want to make any promises yet. From what I’ve seen, Apple is still behind some other AI providers in certain areas, but I’ll test what’s actually possible and see if there’s a good way to use it in Nexy.
thanks for your reply
thanks for looking into this. as another note, Nexi seems to be high in system resources. my fan was at maximum, and my mac got extremely hot and took a restart to fix it. either that, or the amount of Safiri windows i had open, really not entirely sure. I find, generally, AI tasks take a lot of system resources.
I'd like to know
Thanks for letting me know. I’m actually curious about it.
Most of my development and testing has been on an M4. I also tested Nexy on an M1 MacBook Pro with 16 GB of RAM, and it worked fine there. That’s why it’s a little difficult for me to say what caused the high resource usage in your case.
In general, I’d recommend at least an M1 or newer for Nexy.
Could you tell me exactly which processor your Mac has and how much RAM? If possible, I’d like to test Nexy on a similar configuration and see whether I can reproduce the same behavior.