Show HN: Voxos.ai – An Open-Source Desktop Voice Assistant (gitlab.com)
Voxos is an open-source desktop voice assistant that aims to put Clippy to shame while supporting new desktop workflows powered by LLMs.
Tired of copy and pasting ChatGPT responses between your web browser and IDE?
Does your copilot not quite do what you need it to do?
I invite you to give Voxos a try and maybe even become a contributor!
61 comments
[ 50.9 ms ] story [ 1823 ms ] threadedit: shouldn't be hard to enable though
https://gitlab.com/literally-useful/voxos/-/blob/dev/.env?re...
I don't have any direct experience with it... I've only played around with whisper locally, using scripts.
https://github.com/BerriAI/litellm
The post title as written seems intentionally deceptive.
Especially when that something is promoting the opposite of open source. (That's "open sores", which is different.)
Terms like “Recording” make it seem confusing.
Have the responses open in Notepad is confusing in normal work flows.
How does Voxos help avoid copying & pasting code into your IDE? I had a look around the code base and don't see any indication that it allows GPT to directly edit your source files. But maybe I am missing it?
I'm asking because this is a major focus of my open source AI coding project aider [0]. I always like to see how other projects approach the challenge of letting GPT edit existing code. Most recently, aider adopted unified diffs as the GPT 4 Turbo code editing format [1].
[0] https://github.com/paul-gauthier/aider
[1] https://aider.chat/docs/unified-diffs.html
You're correct that Voxos in its current form does not directly work with the user's file system. I'll admit I chose my words carefully in saying that it spares you from copy and pasting between ChatGPT and your IDE - not necessarily that you won't be copy and pasting any more. I feel like having the text response dump to a text editor helps me speed up my workflow considerably when contrasted with the ChatGPT UI being "read-only" in this sense.
Anyways, I'd been messing around with function calling in an earlier version of Voxos and plan on bringing all that work into this beta soon. In terms of my approach, I plan on using docker to host a network mapped drive on the host machine. Then connecting the IDE from the host to the network mapped drive. I'm not sure how well that will carry over to the non-beta version of Voxos that I envision will come with an installer for non-technical users. I haven't put that much work into the idea yet.
An alternative was to host all of it in the cloud and simply offer a web IDE to a container, then make sure there's reliable backup and revert system in place if/when things go south. That's heading more towardsa hosted solution though and I simply don't have time to support paying customers even once Voxos matures to the point I'd liked for a v1.0.
I'll take a closer look at the unified-diffs when I get a chance!
so not gpt4.5-turbo? that's the chat API after all.
While the license of the project may be FLOSS, if all an application is doing is grabbing locally generated input, sending it off to a proprietary third party black box, and then processing the output, the application is primarily glue and the hard work is being done by the proprietary API.
Thus the application adheres to the license of Free/Open Source Software, but not the spirit or motivation, which in this case is primarily independence from third party access to our private data.
By way of analogy, it would be like advertising an electric car that requires a diesel fuel generator inside to charge the battery. Yes, the car may be electric, but is entirely dependent on a machine that only works via fossil fuels.
I've seen a good half dozen "Open Source" applications which work this way and each time it's a bit of a let down.
"if all an application is doing is grabbing locally generated input" demonstrates how little time you took to look over the repo, but feel free to be as offensive as you'd like.
The application was built to be extensible, support local LLMs in the near future - when they're anywhere near as good as the black boxes that are readily available. It's a matter of practicality and it's apparently something lost on many.
You're not wrong in your definition, but neither are we—there's a reason why F-Droid provides warning banners on apps that rely on proprietary APIs. A large portion of the free software movement would rather use an inferior free solution than one that is better but proprietary.
That said, don't take the criticisms too much to heart—I think it's great that people are building things at all points on the free software spectrum.
As for the issue of weather it's "lost on many", I can't speak for others, but I use proprietary LLMs for certain jobs- but I also understand the risk that I place on myself and others when I do so. On the other side, Voxos.ai's documentation's first section is entitled "Voxos [Beta Release] - Voice Your Desires"- voicing my desires would be to give them to a third party. It's a bad idea.
And instead of simply acknowledging the issues and saying "Yes, we have plans for local LLMs" or "We're looking at a mix of rule based and LLM based systems in the future" or even saying nothing, your response is a mix of ridicule and derision. It's very telling.
It actually works somewhat well. I think with some more work and thought, something like this, could actually be useful.
Just saw this was for linux and Windows only
So weird that this isn't local LLM first.