I finally got around to trying this out and I love what you are doing. The teacher I found hard to get to react to the handy but I like the challenge. I attempted to make the voice dialogue to work but it didn’t seem to work, but I see you have it as a WIP so it might not be ready. I wasn’t sure if there was a way to make the strokes change more frequently besides interacting with her each time, but it would be nice to find a way for her to create dialogue without typing to her, but I guess that is how AI Chats work lol. Keep going with this, I look forward to getting some uncensored AI characters.
I am actively involved in both the use of AI for enterprise (we have over 80 AI/ML engines). And since I have access to NVIDIA because of the partnership, might have toys to play with. I am the the one responsible for making our own internal tools secure but also for customers and consumers, I am willing to help to show my gratitude for all the talented people that script and share here. LMK, the LLMs are all similar, some update well, some cause more bus than they fix, .
Better yet, I have the attack surface tools to make sure arent leaving this wide open
I can make it public in a bit, maybe I will make it not so obvious what its for (I might call it a calculator app lol).
The code is relatively complex and even with comments its hard to follow, unless people were working on it in private it didn’t seem like people were working on it so I made it private as it wasn’t worth the exposure on my personal github at the time.
The most efficient way to contribute in my opinion would be feedback on features you’d like to see, as Im familiar with the code and can navigate quickly. I might be bogged down trying to integrate code im not familiar with but if its straightforward I’d be happy to look over it. Im also not opposed to people using it to build interesting projects that might not fit within this framework.
Regarding security, I admit I am not familiar with the ins and outs of that world and you’re probably right. Lets see how it goes, I am planning on adding uncensored models which would be paid on the platform. I will personally use it myself so I want to keep it cheap and i’d imagine others would be interested so I’d like to make it secure, transparency in the code would be wise. I will reach out if I have questions.
Mancer seems cool, but $4 usd per million tokens is steep. This progam uses gpt-4o-mini which is $.15 per million, and I can talk to multiple chat bots for +2 hours and my cost is less than 1-2 cents. I experimented with gpt-4 and 1 hour cost $1.5 which i find excessive.
Let me know if you have any ideas on features, code, or cheap way of integrating uncensored models on the site. I’ll start working more on this over the weekend.
Edit:
It might be worth mentioning that im not from a CS background and not too familiar with collaborative development. If you have resources on how colloborative projects function, I’d be happy to learn more about it.
Hey, super stoked that you got this to work!
I actually haven’t updated the other default characters (teacher, therapist) so im surprised that even worked for you. I will update them tonight.
I reccomend using the character builder it is designed to generate unique characters that know how to operate the handy. I admit im having trouble making the UI intuitive without breaking stuff but heres the gist:
-Select the character builder
-Tell it to generate a character based on anything you can imagine:
Example: “Create a hot female teacher character that’s caught me sneaking out of detention”
-Copy the output
-Click add new character (on the right), paste prompt, add a name, and finally click add character.
Now you can select this new character and chat with it
Honeslty I think the character from the characterbuilder are better and I forgot about the default.
Regarding the audio, I’d reccomend changing the browser as its based on that. It works for me on chrome. I actually prefer the audio over the text because of what you said.
The audio is hands free, you can keep talking back and forth until you’re done, when you talk it interrupts the llm so it feels more like a natural conversation, the functions activates when she says the functions out loud so it feels more immersive. I honestly just use the audio myself, if you want to use the character builer for audio you have to use the one that is on the audio page, not the chat page. I’ll update the UI so it looks more like OpenAI audio chat too. The strokes should change more frequently with the audio.
Thanks for the feedback I’m thinking of adding a characters page with a greater selection of censored and uncensored characters with differnt functions. There could be charactes who doesn’t take user input and simply trigger the api once its done talking so its like a generative JOI character.
Let me know if you have any ideas on features that might be cool
I didn’t think about trying chrome, I usually use Firefox so I will give that a try. One recommendation I would add at this time would be to add the ability to upload some images that can be randomized during the conversation. I guess the easy way around this would be to just put images in a slideshow using a different program, but it might be nice to be able to upload a series of images for each character before you begin and it rotate between them while the conversation continues. Thanks again.
Thats pretty cool, that should be easy add, ill look into it. A bit more complex but i could also tie the images to the functions them selves. This will probably be after I set up accounts and people could configure their long term characters in settings with images.
If visual stimulus is what you’re after, I made a program where the functions activates a video and a script instead of just the motion.
I dont think people really used it but i do so i hid it in the UI to avoid clutter
Click “Chat with Handy-LLM” on the top left 3 times and itll open. Ui is not as nice but it works.
It worked really well! Might I suggest parameter adjustment for Handy range? Some commands would go higher than I could reach.
I did get the voice activation to work last night by switching web browsers. By chance, any way to make it sound more feminine? All I could get was a male voice.
It’s cool you put sort of a backdoor in with the video and script player. I will play around with that and see how it functions.
Hi, wanted to try it, but not working for me (i added my API key), if anyone was able to run it, can you please give a more detailed step-by-step description?
Just double checking, are you using the free trial version of OpenAI or did you pay? The program uses gpt-4o-mini. Without knowing better I tried to use the trial version at first and it didn’t work until I got the paid one.
hi, im using the paid one.
if only it used ais like wwhats on here janitorai.com i wanna find a waycto do it caus no offense all the ais if foud basically all 2 of them kinda suck
same here
Hey sorry got caught up with Exams, I made the repo public again. I made some updates too (added settings), I made some good progress on the uncensored models, and im thinking ill have it working in a couple weeks for other people to use.
Just doing this for fun, ill make a discord later so interested parties can follow the updates.
Edit:
l made a discord, ill post updates here:
OpenAI discontinued the free trial api unfortunately so a minor investment is necessary to get this working
I will post updates on this discord when i start coding, feel free to make suggestions in the meantime:
Hey glad you got it to work, sorry I wasnt able to help out earlier. Feel free to dm me if something comes up.
Hey, I added a settings option (nav bar to the right) where you can tune parameters and test each functions (). Im working on a way to have premade characters with their own unique functions that people can import and tune also, so hopefully soon there will be different functions catering to different tastes. Ill also try to tie quality scripts to functions instead of what I have now.
Ill post minor updates on the discord from now, feel free to let me know of any other features
Hey I tried replicating the male voice but mine were all female. I think its a setting in your browser that might be causing it. I added an additional filter to select female voices but cant really test it since im not getting any male voices. Hopefully it works now though
Also im going to try to post updates and take in feedback on the discord, would be great to have you on.
Being able to connect to Silly Tavern to use the features to allow others AI’s like Claude or Mistral while also connecting the handy would be amazing. It would allow SillyTavern to use jailbreaks so you can have really wild uncensored AI’s…
Still getting a male voice only, what browser settings do you think I could change? I’m using Chrome.