#Plug and Play: Build a G-Assist Plug-in Hackathon!
160 messages Β· Page 1 of 1 (latest)
Hello people. Whishing all participants best of luck! π
Hey, would you be able to clarify which laptop models people get? And can it be exchanged for a cash prize if for instance one wanted to upgrade a pre existing rig?
I'm assuming the 2nd and 3rd place prizes are also 50 series laptops?
Heya. Only the 1st place prize is a laptop, and itβs powered by an NVIDIA GeForce RTX 5090.
2nd place is a GeForce RTX 5080 GPU.
3rd place is a GeForce RTX GeForce RTX 5070.
Thanks for the clarification π
And about whether they can be exchanged for cash if we should want to upgrade a different part. Or say I preferred an A6000 for my more professional applications?
Also which laptop is it?
π 5090 GPU Laptop .... will this laptop has same 32GB VRAM as 5090 GPU. if the laptop has same horse power as 5090 GPU, there is very cool.
It doesn't have the same horsepower or vram as the desktop counterpart, not even close. And I want to know, is it a Legion Pro 7i, or a ROG, or what? I need the model so I can see what else is in the pack
This competition entry requirement is not low as G-assist requires 12GB+ VRAM GPU. My only available GPU is Jetson Orin Nano Dev Kit π€¦ββοΈ
No HW to test G-Assist ....
@zenith condor ?
Sure thing! Thanks for the patience with the slow reply, was out of office for a few days. But some answers for you:
Prizing cannot be exchanged for cash, and there are no customization options for prizing.
As for the laptop, its a Razer Blade 16β with a GeForce RTX 5090. https://www.razer.com/gaming-laptops/Razer-Blade-16/RZ09-05289EN4-R3U1?srsltid=AfmBOorBWT9hbb0oStBMh0it3cyviGP3g2QUKGf-vIGuZrhbJxWvIIUp
Purchase Razer Blade 16 - QHD+ 240 Hz OLED - GeForce RTX 5090 - Black (gaming laptops), or browse through our extensive selection of gaming peripherals, and more at razer.com
Wow .... this notebook is awesome π
Nice laptop, I suppose one can always sell the desktop GPUs π
Thank you very much for the response
@zenith condor I read the Hackathon terms in pdf file. I would like to use G-Assist to develop useful plugin which is not for gaming but another non-gaming meaningful and useful function G-Assist. Can I do this? Or the judging is solely for gaming related?
Captured from PDF file
The Challenge
This contest involves the following challenge:
- Build a G-Assist plugin that helps RTX users control a broad range of PC settings, from
optimizing game and system settings
Can you share your plug-in idea with me? Youβre welcome to DM it to me if youβd like
Thanks Candra. I sent DM.
Just following up here for everyone -- we're excited to see all your ideas for the Hackathon and what plug-ins sound like good ideas to you!
Happy to help with specific ideas and questions about your projects. But to clarify, entries do not have to be related to gaming. They can be about any PC workflow you find interesting or would help you and others π
I cant get acces to G-assist with Nvidia App 'cause my RTX 3080 got 10Gb instead of 12 ?!
Did i need to spoof the app or i cant try g-assist on other way ?
i cant make "gassist run" with you git REPO, so...
Hey @zenith condor what do I do if I don't use social media? And also what to do with the fact that my 4060 has on 8GB of VRAM? Am I not allowed to compete, or is it just the app won't run, or will run very poorly without the VRAM?
alright, I think I see an answer in another thread. I can build a plugin, but not test it without a 12GB card?
Ok, g-assist found, let's play with sura...
Hey Serda, just checking in -- you were able to get everything to work?
to be fair, i grabbed the nvtopps folder on a compatible g-assist systeme, and i try with the standalone sura, for the moment, it's work π
What kind of things everyone working on?
Working on how to get the G-Assist on. I cannot meet the system requirements π€¦π»ββοΈ
i got the g-assist from a compatible system to test it, idk if i can share this, or dev can give us something to try π
but u can put this : https://github.com/NVIDIA/G-Assist/tree/main/api/bindings/python
& plug llama3 to try & simulate g-assist β€οΈ (llama3-7B 8-10Gb of VRAM)
@zenith condor can i dm ? got few question about plugin and idea :p
Sure, feel free to DM
Has anyone else encountered Plugin {plugin_name} took too long to respond.? I'm getting this, but I'm seeing a successful response in my plugin log file. I'm not sure how to debug this.
Fixed! β
I was missing <<END>> at the end of my JSON message
Done β€οΈ
Will the "RTX AI Workshop: How to Build A G-Assist Plugin" video available on YouTube?
I have a tech question. I followed the instruction to run setup and build the exe from examples/weather. I got the weather plugin exe file. But I tried to pass input data to it but the exe does not provide the output to me. May I know if I can verify the exe along without having G-Assist integration. I would like to verify and isolate the problem. This is the example code from examples folder /weather.
Thank you for your help in advance.
did u try something like this ?
echo '{"tool_calls":[{"func":"shutdown"}],"messages":[{"role":"system","content":"Close the plugin."},{"role":"user","content":"Shutdown."}]}' | .\dist\g-assist-plugin.exe
Just tried. I wonder if my build exe having problem or not. The exe halt and I have to clt-c to break it.
C:\code\g-assist-hackathon\G-Assist\plugins\examples\weather>echo '{"tool_calls":[{"func":"shutdown"}],"messages":[{"role":"system","content":"Close the plugin."},{"role":"user","content":"Shutdown."}]}' | .\dist\weather\weather-plugin.exe
Traceback (most recent call last):
File "plugin.py", line 270, in <module>
main()
File "plugin.py", line 163, in main
command = read_command()
^^^^^^^^^^^^^^
File "plugin.py", line 216, in read_command
logging.error('Error reading from command pipe')
File "logging_init_.py", line 2115, in error
File "logging_init_.py", line 1518, in error
KeyboardInterrupt
[PYI-9532:ERROR] Failed to execute script 'plugin' due to unhandled exception!
for the exemple:
`def shutdown(params=None, context=None, system_info=None):
return {"success": True, "message": "shutdown"}
COMMANDS = {
"shutdown": shutdown,
# other command
}
SHUTDOWN_COMMAND = "shutdown"
cmd = ""
while cmd != SHUTDOWN_COMMAND:
input_data = read_command()
cmd = tool_call.get("func")
func = COMMANDS.get(cmd)
if func:
response = func(params, context, system_info)
write_response(response)`
and, done, command shutdown working β€οΈ
bonus tips, if you want ctrl+c to be clean:
except KeyboardInterrupt: logging.info('Plugin stopped by user (KeyboardInterrupt). Exiting cleanly.') sys.exit(0)
β€οΈ
Really appreciated to your help π
Let me try it again
No problem β€οΈ
Yes! It will be π
Interesting question on submission form 13. What NVIDIA hardware did you test your plug-in on? * π
Hi everyone! I'm here with the G-Assist engineering team, and we'd love to help answer any questions you have about your G-Assist Plug-in hackathon submissions. We'll be in the channel until 2pm PT or so. Feel free to @ me in your questions and I'll make sure we get the right people looking.
To kick things off, we had a few questsions from last week's workshop we didn't get a chance to answer. Sharing those below:
what is the best way to test your plug in if your laptop dosent meet the g-assist system requirements?
It's hard to test the plugin if you can't run G-Assist, sure. Maybe a friend can help. You can find system requirements for G-Assist here: https://www.nvidia.com/en-us/software/nvidia-app/g-assist/
That said, laptop support is coming soon. π
Does G-Assist support multi-gpu on the backend? Ex: If I have multiple GPUs in a target system, will they be utilized? *If this is supported, with things like the LLM, and ASR models run on run concurrently on the different GPUs?
Multi-GPU is not currently supported by G-Assist. But we appreciate the feedback.
Hey @forest dome ! I was trying to install the g-assist app so i can try my plugin but i ended without any sucess. Can you guide me how to install(i have an 4090)
Btw i have submited the plugin as it was π€£
I just completed. And I am thinking to submit it today or submit it on the deadline date.
I worry is my project will be cloned and submitted by another ppl π
Installing G-Assist is pretty easy.
- Download NVIDIA App.
- On the Home page in the Discover section, there should be a card with G-Assist on it.
- Click the GET button and the download should start
Glad to hear it's done! We can't wait to see what you submit. π
That is a G-Assist Plugin π for non-gaming use.
it just doesn't appear, dont know if i made something wrong
You're on desktop?
hey there folks , i have a question :
- i want to build basically a plugin that changes the keyboard color based on financial news / releases + runs a model called amazon/chronos for forcasting , but i'm afraid the "theme" is kind of adultish , meaning is like financial time series forcasting for equities actually palatable to the competition ? my main idea is that someone that's deaf couple basically get cool alerts , but i dont really want to be "that guy" and spend a lot of time on it if it's out of the question sort-of-thing
I spent almost a day to prepare the video with the voice over. Even that is AI TTS. It takes time.
laptop
That's why. G-Assist does not yet support laptops. Laptop support is coming soon.
i thought that was the whole idea lol , far out far out
lol, I ended up sending it without testing anyway, thx for your attention π
The G-Assist Plug-in Hackathon doesn't have any restrictions in terms of plugin "theme." Non-gaming use cases are fine.
but i want to win π hopefully i can π fingers crossed , thanks for the advice
Appreciate all these questions so far. Keep 'em coming! 
@forest dome do all the submissions get a feedback?
Not automatically, no. But we're here to give you feedback now if you'd like.
Will the results be public? Once posted, can we ask for feedback?
The winners will be announced as part of the contest. In terms of feedback, you can always share your project here for feedback -- not just from us, but from the community too.
That's why this discord's here. π
Also, can you give us a spoiler about what else Nvidia is building for us?
I cannot. π
i hope someone make a plugin to order pizza from g-assist
"hey g-assist, if my gpu reaches X percentage of usage, order a peperoni pizza form me"
This is not difficult. I believe you can make it. And then, share it π
This is just an API call ....
The challenging part is having enough money in your wallet ...
so, what if we connect with a mcp with a LOT of tooling for every simple thing ever?
order food, items, lock dors and etc...
for be honest the first time i saw the hackaton page i trough that any plugin could be maked, even with microphone inputs...
started to make a realtime audio censoring but i gave up
Why'd you give up? Just curious.
lol last question because i've been super busy (same as everyone?) and folks are asking for feedback : when is the deadline ?
5:00 PM Pacific Standard Time on July 16, 2025
wish me luck π π
Good luck!
and good luck to all the submitters ! btw if someone (competent) with interfaces wants to help me make the trading news plugin , lmk π
We had one more question from last week's workshop that we have an answer for. Just sharing for everyone:
I'm making a few network requests in my function that take some time to finish, and sometimes I'm seeing a timeout (Plugin took too long to respond). Is there currently support for partial updates to avoid the timeouts I'm experiencing? Are there examples of this? I saw some related code in the rise python bindings
Yes, partial updates are supported. You can send a JSON response via the ""write_response"" function; just don't include the ""success"" field. There is a 10 second timeout we allot for plugin responses, so make sure you send them before that times out. Every response resets the timeout.
Example:
write_response({""message"": ""Still processing network request...""})
via dm (i'll be getting this done next two days)
Thank you for answering this one @forest dome ! (that was my question!)
For now I have a workaround where I use threading for long running tasks and I'm able to return immediately. That makes sense, good to know about the timeout reset! π
Glad we could help!
lots of things to do in a short space of time (I started Saturday night)
Maybe I'll opensource this, I've done some things but there are still many more to do, echo filter, improvement in the audio buffer for transcriptions, UX, latency improvement (I'm getting approximately 2s of latency initially)
Hi y'all, the team's gonna have to move on to some other meetings, but we really appreciate all the questions and comments here.
Please keep the questions coming. I'll share them with our engineers and we'll do our best to help. Thanks everyone!
will us laptop users get a chance at winning prizes in like a second competition?
push it, and let see π
Hi all, the web page now showing "from June 17 to July 20." Has the deadline changed?
I also assume that the deadline has changed; the date of the 20th also appears in the terms and conditions.
This is a good thingβdeveloping an effective and relevant plugin is not as simple as expected!
https://developer.nvidia.com/g-assist-hackathon/submit this link is not working for me (chrome:latest , windows11 , paris, france)
aaaand now it works totally fine , which is surprising and nice
Submission done! Letβs goooo π₯
Yes, its now the 20th. Glad to hear more time is helpful! Excited to see what you develop π
Sweet, glad its working!
Just submitted π
@lilac yoke
@forest dome
Hell yeah !!! try to finish it, and push it !! let's goooo !
Hi @forest dome, I'm following up on my issue with the plugin response timeouts. I was able to send intermediate progress updates by leaving out the success: True parameter in write_response, but it looks like I'm still getting stuck at the 10 second timeout.
The message I'm getting says: Plugin {X} took too long to complete its response. I'm doing local image inference with the flux NIM, and the request takes about 20~30 seconds to complete. Is there any trick to getting the timeout counter to reset within a single request to G-Assist? I might have misunderstood how the timeout reset works.
Hmm, interesting. Let me take this to the team and see if we can figure it out.
@forest dome
I'm also going to assume therefore, as I am on a laptop, there is no point in competing
Unfortunately G-Assist doesn't currently support laptop GPUs. That support is coming soon, but you'd have a hard time testing unless you have a friend who has a desktop system you can try it on.
how many entires do you already have?
@zenith condor Hi! Can G-Assist plugins draw on NVIDIA Overlay?
Developing a G-Assist plugin that requires visual feedback. Is there an API to render custom graphics
On the NVIDIA Overlay (where the FPS counter appears), instead of creating separate overlay windows? The current plugin manifest only supports command definitions, not UI elements. Any guidance appreciated!
No, G-Assist can't draw onthe Overlay at this time.
i need a friend for this to build my submission without breaking the bank on cloud , does anyone have a few minutes to help me out with a remote or even to run the build commands on their end and push ?
we can cut the main prize in half with a saw if we win np π
πͺ
ho boi... what a gift !
this damned <<END>> , lmao π€£
hey hey hey , so i'm wrapping up and submitting but basically the app is like 17GB lol so that's not really going to make it on my free github accout, hopefully i can share the spec and instructions in the readme
the app itself is still less than 10K loc but basically when it gets wrapped up the plotting and databases get quite heavy, also there's all of torch , triton , transformers and full model(s) in there lol, but at the end of the day that's the tradeoff you make for a fully local app π i saved a lot of space just using the default nvidia sla model i think , so there is that π
Just submitted LoreMaster for the #AIonRTXHackathon.
Really glad the deadline was extended β I only found the hackathon on the last day and built this in under 24 hours.
I hope the source code helps other last-minute folks cross the finish line too. Good luck!
GitHub: https://github.com/mwtuni/loremaster
Demo video: https://youtu.be/gzQvNmVxp_8
a GβAssist plugin that lets you talk to ANY game character in ANY game real time via voice or chatting. Built in a day! - mwtuni/loremaster
LoreMaster is a custom plugin for NVIDIA GβAssist that lets you talk directly to your favorite game characters β like Cloud, Aerith, and Tifa β using natural language, voice, and GPT-4o.
Ask lore questions, request in-game help, or just roast them about their oversized weapons.
Features: Voice queries and responses, accurate game info, s...
I had a similar issue with my LoreMaster plugin (as rendering the speech took longer time than allowed). I resolved it by playing speech in a thread. See my source code for example, posted on this channel
...also in your manifest.json you should set persistent to true (for the timeout issue)
nice, yes I think that's a good way to go for tasks that take longer. the flux NIM that I used in my plugin takes about 30 seconds to generate images so I also used the threading pattern π
Congratulations to all submissions ππ
My 1-day LoreMaster plugin is still quite young β built it just in time for submission β but Iβm already adding some killer features post-hackathon: a hybrid local LLM setup for private conversations and a VLM (llava) for game context.
To avoid interfering with judging, Iβve frozen the main branch to reflect the original submission and will continue development in a separate dev branch.
Is this the common practice in hackathons β freezing main and continuing on a dev branch post-submission? Curious how others are handling it.
I don't know why I tried to run weather plugin example in G-Assist, it has no response. Does anyone has idea?
Typical G-assist function are working good.
I guess that is some permission problem about G-Assist but I have no clue where can I find the issue or enable G-Assist debug info.
Not the plugin issue as I just test the example from G-Assist folder in my newly setup Desktop.
log ? π
No luck. I sent email to NVIDIA to get help π€¦ββοΈ
damn....
I would recommend to check "list plugins":
@daring apex saw your email, we're looking into this.
Thanks Cory π
Thanks mwtuni π
List provider is OK and showed the list correctly.
I found out the problem finally. G-Assist plugin cannot work with Remote Desktop.
I setup a new Windows to test and I am sitting in front of the Physical Desktop. I run the example. it works.
Then, I enabled the Remote Desktop accessing remotely. Not working.
I tested it in front of the desktop again. It works. π
omg who one this one i wonder, it was a tough one for the build !
We haven't announced winners quite yet. There's a bunch of good options. π
That said, we did just launch our new plugin portal with mod.io. We'd love to have you add your hackathon submissions here! https://mod.io/g/g-assist
https://mod.io/g/g-assist/m/aria-avatar-companion I forgot to add .zip file to github release page. is that counts?
Any result yet π
They're too busy running all those awesome plugins
Keep chatting with Aria and generating image π
Jailbreaking Aria to be a soulmate
@forest dome any updates on the results? Thanks!
No updates yet. We have finalists and I need to reach out to them. I'm on it, y'all. π
i havent recieved my victory email yet, so dont be shy to just send it π
lol!
any update?
@forest dome any updates on the results? Thanks!
Hello, any updates please?
@forest dome it's been a while though, huh ? let's lock this in ?
#rtx-ai-pc-announcements message
heartbreaking + congrats to the winners !
congrats!
Hi, I'm working on a plugin for the home automation hackathon. Idk if this is the right place to ask programming question, but here goes. When the plugin gets a query it check if any initial setup needs to be done and if so executes that. I have "persistent" in mainifest.json set to true. I assumed that setting that to true, means the process stays alive and thus any subsequent requests will still be in the same process meaning that all state in the RAM is still there. Instead I see that the main function is called every time.
^ I'm also wondering this.
Also, does the manifestVersion do anything? Will g-assist not pull in my updated manifest unless I bump the manifestVersion, or will it load on startup every time regardless?
On my platform, the initialize method function is never called so I do a check anytime the script starts. The shutdown command is called. The "persistent" setting is respected which means that the host keeps the plugin process alive and it reads commands through a loop.
I haven't messed with manifest version but I have to restart g-assist to clear out the persistent plugin process in between updates.