7GB is not too much to be honest.
Some script videos are double that size 
I just hope it won’t requite a fancy GPU fingers crossed ^^
But really, curious to see where that goes 
im having issues with no responses past the inital message in Vanilla too, not sure why
still gonna be opensource ?
@yankees155 = VanillaGPT is stand alone for most part. The zipfolder itself contains a userguide on what is still required to run it. Make sure no files from the zip are getting deleted by an AV or something. Unfortunately, since Vanilla has its generic source code released it is no longer an active development thingy so you will have to resolve issues on your own.
@hotncoldi
I will be posting it in this thread(which is based on an open source project)
So nothing has really changed. It will follow Vanilla’s path. A working prototype 1st then later the generic source code
(when i am 100% sure its not gonna be used for something new / when i am fully done with it.).
3 Likes
thanks, yeah im not very ai savy so idek. tried reinstalling everything and the responses are suuuuper slow and seem to keep resetting the convo
I wanna try this so badly but the set up and installation is so tedious. I hate it…
@yankees155 Yes, I tried to make it semi-plug and play cause binding an entire AI model for offline local deployment did not seem feasible considering the sizes. Conversations reset because they are treated as a new clean session(meeting 1st time). Vanilla was the 1st prototype to figure things out. You would need to elaborate what is super slow while running the prototype to determine what could be the bottleneck.
@tsuriley most of these opensource projects don’t natively have a wrapper. While there are some that took this concept and turned it into an actual web powered platform, Vanilla and Misa are more meant as local prototypes rather then a proper product with less setup.But unlike Vanilla, Misa has an dedicated startup menu with colored indicators. You cant start the actual application if those arent green (a fail safe to prevent errors later on). But i never really posted previews. Determine for yourselves if its worth the download lateron nods
1 Like
I guess my point is that if this setup is still the same, because I found it waste of time and space for this when it first a thing. My only problem is following the instructions. I don’t know if this set up is easier then what it used to be.
Hm, how should i put it. The prototype version are created in such a way you don’t need to worry about most of the backend stuff like booting up things like python or entering commands. In particular Misa would work like this: Download the zip/folder ----> launch the Exe. -----> start up menu has buttons you click to install additional requirements. When all visual indicators are green you can launch the actual App.(it will never allow a launch unless all required checkpoints are cleared 100%) The original StrokeGPT does not have that and is purely a raw concept.
If you meant if the fundamental concept is still the same, then yes. it is still a Handy only prototype. But its for that very reason that its posted in here and not as it’s own thing.
I appreciate the help, this is where it gets stuck, seems to keep repeating and stops recognizing my responses, just times out and kind of resends the same starting message and stops adjusting the mood meter, its ok though im fine with waiting for misa!
This is my latest attempt to rework the StrokeGPT idea. As far as I know it’s the only actually usable version of this idea at the moment. Assuming your hardware can run it fast enough.
Vanilla was not really build with deep intricate conversations in mind. (it was exploring other elements such as alternative methods of strokes sync to visuals etc)
It also has no guard rails for input spam. The user would best wait with a response until Vanilla is completely done with her own reply (chat message + thought bubble).
What would also break the prototype is not removing the old server from Taskmanager processes 1st before starting a fresh boot.
The prototype expects the User’s own (RP) push for sexual/Physical Escalation.
e.g: User might start casual, but can ‘‘nudge’’ Vanilla into that NSFW direction very fast: chatting with Vanilla is but a formality to have ‘‘naughty fun with her’’. The Pussy vision/progress bar are visual indicators for user to be more ‘‘Bold/Direct’’. How it usually goes: User can start casual (hows work/ its good to see you again, etc) ----> Amplify what she tells the User with something lewd ----> Increase the intensity to push the progress bar fully into Escalation ----> When Vanilla presents the special question(Pink label) respond with Yes or No only.
Misa will probaly work a bit better in that regard and some more. I am pretty close to finishing it. Just some more testing its local/stand alone nature.
https://www.youtube.com/playlist?list=PLBhXxyutbCwQ
Here’s some videos to walk you through installing and using it if you need help.
1 Like
thanks! looking foward to it
Where IS the Download Link?
Its clear you put a lot of effort into making a very complete and versatile prototype. You should post a demo video showing its capabilities in actual practice a bit.(little chat session, etc) A lot of peeps are very much motivated by visual stimuli and less much by technicalities (even if that’s ofcourse an important element).
Do you mean showing it on the actual device/a penis? I’m not going to do that but if there’s any volunteers, I might take one, lol (me recording the interface on my end, you recording the changes on yours).
If you mean like a non-robotic conversation with motion status changes, I can. It’s low on my list of priorities though.
1 Like
I don’t think either GPT or Claude would be willing to do a proper demonstration with erotic language. But I could try when I redo the guide with the new UI that I just pushed out.
I’ve gotten this to run on a Mac. I don’t have a handy (well its somewhere and most of my focus is on the SR6 and a motion mapping software) so I am currently trying to figure out how to get T-code to work. The software currently just asks for a handy key before anything so I can’t really inspect how well it works on MacOS.
I know it is open source so I’m probably gonna modify this code to try and suit an SR6 - or at least talk to it on the single main linear axis.
Sorry if this has been done but the comment section is long and I didn’t really feel like reading through all of them and just decided to see if I could get it to run on a Macbook.
The single biggret setback that I can see was which port you were using. Port 5000 is generally occupied on a Macintosh so I switched it to 5001 and that let me run the software (after downloading Ollama and all that jazz) … Working on a full workflow for Mac users, so I can at least give back a little to the community until I get this tracking software up to a usable standard.
“app.run(host=‘127.0.0.1’, port=5001, debug=False)” was what I changed from 5000. in app.py
WHAT YOU NEED
• macOS Sonoma 14 or newer for the current Ollama app.
• Python 3: Python Releases for macOS | Python.org
• Ollama for Mac: Download Ollama on macOS
• An internet connection for initial downloads, Handy cloud control, and optional
ElevenLabs voice. The language model runs locally through Ollama.
• Apple Silicon uses GPU acceleration; Intel Macs use CPU only with Ollama.
Model performance depends on available memory and other running apps.
-
INSTALL OLLAMA
Download the Mac app, open the .dmg, and drag Ollama into Applications.
Open Ollama. Accept the prompt to install its Terminal command if shown.
A background activity notification is normal. It may run without a visible window.
If a script says it cannot find the app by name, try in Terminal:open /Applications/Ollama.app
If that path does not exist, install from the .dmg
-
DOWNLOAD THE MODEL
Open Terminal (Command-Space, type Terminal, press Return) and run:ollama pull llama3:8b-instruct-q4_K_M
Wait for “success.” Keep Ollama running when using StrokeGPT.
This model name matches the original project’s default configuration.
Open http://127.0.0.1:5001 in a browser. Keep Terminal open.
If macOS will not open the launcher, open Terminal and use the manual method:
Type “cd ” (including the space), drag the extracted folder into Terminal,
and press Return. Then run these lines one at a time:
python3 -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements-mac.txt
python app.py
If the launcher lacks execute permission, from that same folder run:
chmod +x Start-Mac.command
./Start-Mac.command
I mean it works… At least the chat part of it.. I haven’t hooked up a machine to it. The software is memory dense.