This looks very amazing. Thanks for sharing this amazing build jesús. Thats so nice of you to share this with complete codebase. A lot of people tend to gatekeepe it to them only. Keep up buddy. Waiting for more vids like these. And ill also be keeping a tab on GitHub repo if there are future updates. 🙌
Imagine for a moment the potential of Samantha in fully digitalized companies with microservices architectures consumed through the tools module. It reminds me of that AI from 'I, Robot' by Isaac Asimov. Great job, keep it up
That looks awesome Jesús ! Would it be possible to use local LLM (with Ollama for example) and local stable diffusion models (using forge webui or comfyui)?
Yess, absolutely, if you're lucky enough to have a beefy GPU for that hehe. You'd need to change the llm instantiation that uses Groq for Ollama -> python.langchain.com/api_reference/ollama/llms/langchain_ollama.llms.OllamaLLM.html#ollamallm
@@jesuscopado-en Thanks for the tps, will look into that :) You don't need a super beefy GPU to run all LLMs some mistral or llama models are useable on more modest configs too!
@@TheYules yeah, you're right! But you gotta stick to less than 7b parameter quantized models which are not that great. I usually test them on LM Studio app on my mac, great app actually! I always find simpler to go to Groq/OpenAI APIs as I don't worry about privacy (yet) and value the better performance
Thank you for sharing this interesting proposal. I would like to ask you what additional resources may be required if you want to use it in a language other than English. For example, Spanish.
I like your work. This is awesome. How can I build this for myself usisng the tools you provided? I watched the other video but still can't figure out the process. Where to start from?
Glad you like it :) Hhmm I'd honestly recommend to ChatGPT your way into understanding. If you're new to Python or programming, ask for the basics, then take screenshots or copy paste errors and ask for clarification, that would both help you learn and progress in the projects.
Great project man. I can't get it to run though. The voice/mic button doesnt appear and i get errors. Do you have a community I can join to trouble shoot? I would love to build on this project!
wow amazing! But I didnt get it. you use flux for image creation which is absolute awesome! Do you run it locally? Im running it in forge on runpod. Do you have an idea, how I can connect that frum runpod to samantha?
Ah dang... Dunno what to tell you. Annoying to not have the error trace yeah. Maybe the endpoint is not correctly set... I'll soon update the repo with the latest Chainlit version (not the alpha) and maybe that helps.
I have seen your TH-cam it's amazing. I have a question can you combine your all videos and make an AI Sementa live video when you asked her she responded that instead of audio make her a video with the expiration from the previous girl that you make emotion picture to video
I'm glad you like my channel :) That'd be complicated, cause the avatar video models and lip sync models are still very slow, but soon in for future for sure this will be possible!
Nice job! II hade an idea talking with my mother. What if we will try to improve and customize it focused on cell use for parent and grandparents to use it in their cells Mostly, don't know how to use a cell. But, with an AI named as they like. I think it will be great for them :)
This looks very amazing. Thanks for sharing this amazing build jesús.
Thats so nice of you to share this with complete codebase. A lot of people tend to gatekeepe it to them only.
Keep up buddy. Waiting for more vids like these. And ill also be keeping a tab on GitHub repo if there are future updates.
🙌
Appreciate that very much!! :)
And I'm with you, sharing is key to progress, long live open-source haha
This is great, exactly what I needed, thankyou was having hard time getting tools with realtime api
This is absolutely incredible!
really cool stuff and i love it is showing examples and straight to the point! Thanks for sharing this!
It's crazy yeah... Glad you liked it :)
wow very impressive !! love your content, and keep it up !
Thanks a lot! :)
Imagine for a moment the potential of Samantha in fully digitalized companies with microservices architectures consumed through the tools module. It reminds me of that AI from 'I, Robot' by Isaac Asimov.
Great job, keep it up
Yeah, the potential of that is huuuge. And great book! :D
So interactive... I love it
I love love love this and thank you for sharing, how did you build your tools ?
Awesome! Would appreciate more insight into the .env setup and what API Keys are required.... for example does it have to be Azure?
it seems that several of the keys are paid sites.
Next Python build video will show entirely to set up the .env from scratch and configure both TogetherAI's FLUX and Groq! Uploading it tomorrow!
Some of them are free till a certain usage, but plenty for a single person. I still haven't paid a cent :D. I mention that in the comments.
The interuptability looks really good, 'she' shuts up quickly when you start talking
Yess, that's crazy good, and contributes a lot to the feeling of having a human-like conversation :D
Awesome!! ❤️
Thanks! :)
Love it! I gonna subscribe, give you a like, watch all the other videos you may have and activate all notifications!
Thanks a lot!! :)
Amazing welldone thanks for the upload. new subscriber here
Appreciate that! :)
That looks awesome Jesús ! Would it be possible to use local LLM (with Ollama for example) and local stable diffusion models (using forge webui or comfyui)?
Yess, absolutely, if you're lucky enough to have a beefy GPU for that hehe. You'd need to change the llm instantiation that uses Groq for Ollama -> python.langchain.com/api_reference/ollama/llms/langchain_ollama.llms.OllamaLLM.html#ollamallm
@@jesuscopado-en Thanks for the tps, will look into that :) You don't need a super beefy GPU to run all LLMs some mistral or llama models are useable on more modest configs too!
@@TheYules yeah, you're right! But you gotta stick to less than 7b parameter quantized models which are not that great. I usually test them on LM Studio app on my mac, great app actually! I always find simpler to go to Groq/OpenAI APIs as I don't worry about privacy (yet) and value the better performance
Thank you for sharing this interesting proposal. I would like to ask you what additional resources may be required if you want to use it in a language other than English. For example, Spanish.
Sii, se puede en español, solo tienes que hablar en español para que te responda en español. Como con ChatGPT
I like your work. This is awesome. How can I build this for myself usisng the tools you provided? I watched the other video but still can't figure out the process. Where to start from?
Glad you like it :)
Hhmm I'd honestly recommend to ChatGPT your way into understanding. If you're new to Python or programming, ask for the basics, then take screenshots or copy paste errors and ask for clarification, that would both help you learn and progress in the projects.
Great project man. I can't get it to run though. The voice/mic button doesnt appear and i get errors. Do you have a community I can join to trouble shoot? I would love to build on this project!
Ah dang, if it's related to the mic try searching in Chainlit open issues -> github.com/Chainlit/chainlit/issues
wow amazing! But I didnt get it. you use flux for image creation which is absolute awesome! Do you run it locally? Im running it in forge on runpod. Do you have an idea, how I can connect that frum runpod to samantha?
It's running through the TogetherAI API. Well, you could pull the repo from runpod and then open a ngrok tunnel to use it from your machine
@@jesuscopado-en ahh thx! good idea!
Hi, getting stuck at Connected to OpenAI realtime: - the mic turns on and nothing else happens, no error codes even. API Keys set.
btw Together now only offers $1 of free credits.
Ah dang... Dunno what to tell you. Annoying to not have the error trace yeah. Maybe the endpoint is not correctly set... I'll soon update the repo with the latest Chainlit version (not the alpha) and maybe that helps.
@@jesuscopado-en Cheers! :D
Looks like gpt-4o-realtime-preview, 2024-10-01 is available only in eastus2 and swedencentral
More subs for Jesús
Appreciate that! :)
I have seen your TH-cam it's amazing. I have a question can you combine your all videos and make an AI Sementa live video when you asked her she responded that instead of audio make her a video with the expiration from the previous girl that you make emotion picture to video
I'm glad you like my channel :)
That'd be complicated, cause the avatar video models and lip sync models are still very slow, but soon in for future for sure this will be possible!
Nice job! II hade an idea talking with my mother. What if we will try to improve and customize it focused on cell use for parent and grandparents to use it in their cells Mostly, don't know how to use a cell. But, with an AI named as they like. I think it will be great for them :)
That's a great idea! And I think it'll come in the near future... Many companies are already betting on this kind of personalized AIs
great, exactly
you look shy when you talk to her, that's cute
hahahaha, wait till I configure her with Scarlett Johansson's voice :D
what's the cost breakdown for this demo?
Zero, running on those free resources and trials I mention in the comments :D
At this time Samantha have memory ? can she remember what you talk in last chat or 3 days ago ?
Noup, not yet, but I'll implement that soon! In a similar fashion of how ChatGPT does it now
Nice jobs ❤
Thanks! :)
cool… i can see the agentic future is here
Completely agree!!
always getting "Failed to connect to OpenAI realtime:"
Ah dang, everybody that I know has been able to replicate it, if you send me the full log trace of the error I can try to help
With this pace, I am sure AI Overtaking Humans era is very near.
Feature Request 01: Please give Scarlett Johansson's voice to it. Pretty pleeeeease.
Absolutely, but that'd be the end of my in-person/human social life haha
imagine claude desktop and smantha ?
Gotta work on that!!
not her, the sound is a bit robotic, if you can work on voice texture this would be surreal
Yeah, definitely doesn't sound like Scarlett Johansson hahaha. But we'll get there!! :D
well, its not going to be completely a human voice. Because it needs to wait and listen then process (NLP).. the movie Her, was just that a movie.
@@mine4me403 i think you haven't yet talked to openai's voice model.