This is great! Do you know what I would do if I want to keep track of what user is calling? I'd want to somehow pass the userId from the UI down to a custom LLM where in the background it will use the userID to bring in the memory from previous interactions.
Thank you for another great video. I really do want to see the details of how you customize the prompt. I have a vapi assistant with a script in the system prompt with branched logic depending on the user response. I would love to move that to a custom Ilm to be better able to control the flow, logic, and responses. I have seen your “conversational pathways” video but it’s just not clicking for me. A better understanding of how the python program and custom Ilm work together - like a flowchart.
You might want to look out for Vapi Workflows - will be released soon, and I believe that'll be super beneficial for your agents in general to have more granular control.
can u integrate Agency Swarm into vapi and make a video showing how to do it? so that agency swarm can have complete control over the business and also do emails and sms etc. call backs and setup appointments ?
I don’t think Vapi charges extra for the OpenAI usage. You might have more granular control of the length of the prompt. For example, if the first message was said, you can change the prompt through a new one that removes the greeting part from it, which causes lower tokens during the call, which in return saves you a bit of money
Thanks Jannis. Does using a custom LLM provide more security against using the VAPI provided models? Instead of VAPI taking the data from transcripts and it going to their model it is stored with ours?
It would certainly cut out one middleman. You can leverage a custom-hosted solution that is hosted based on your standards. However, while it makes it a little more privacy-oriented, it's only as secure as your developers or devops are good.
Is the API key from openai a regular api key? Can you do another video on customLLM but using not only openai key but Assistant api key as well so then we can configure the assistant on openai and just use vapi for the voice. But instructions & functions are all done via openai! Also thank you for doing this video as a beginner! Literally the only thing I need help with is having vapi work using custom LLM connecting it to my assistant api (asst_key)
@@jannismoore Been trying to accomplish this also! Would love an explanation on this.. The endpoints for assistants are different to chat completions...
Thanks a bunch for another great video. Curious, I've used Air, and in Air all we need to provide is a call script. I'm now planning on learning Vapi, but I have no idea how to prompt in Vapi, to get an output similar to my current call script. Can you use call scripts in Vapi?
I think they use a custom Prompt to generate the master prompt including your transcript. I'm not a fan of this as scripts usually have a lot of nuances. It's best to custom prompt engineer them in Vapi for the best outcome.
@@jannismoore thanks. I was "afraid" of this :) I guess I'm better at call scripts than prompting. I'll give it a shot and see if I can achieve similar output through prompting.
Great video, thanks! Can you perhaps make a video on how using AnythingLLM (docker on a small VPS) as a custom llm for VAPI? They have an API endpoint compatible with OpenAI Completions. Somehow I can't get it to work.
It can be done with Laravel, and I doubt there are any tutorials specifically on that available, but here's a prompt that let's you get started with ChatGPT: Create a markdown formatted manual that generates a manual on how to connect Vapi with a Custom LLM Setup using Laravel. Ensure it's easy to understand.
Gotcha. I've got quite a few videos that are important when doing Outbound calling. Check out these: - ruclips.net/video/lktoSWsHbiI/видео.html - ruclips.net/video/3y7GJ9Vozkk/видео.html - ruclips.net/video/iWZD_HBk9zQ/видео.html
Hey man, glad to see you join the space! I've limited my available timeslots, but you're most welcome to join my Discord. You'll find more about that on my channel page. :)
With the example I'm using there's no authentication as it's a proof of concept. The credential endpoint isn't even used. What exactly are you trying to achieve?
@@jannismoore I was referring to the authentication instructions from Vapi's docs on "Using your server" and I've got it working now, but the details in the docs are limited. For example, it doesn't say how to update the api key in the credentials or manage multiple environments.
I can, but with the current setup, I don't recommend it as state management is incredibly difficult if it comes to things like function calling and interrupted messages. I've built it out before and had tons of issues with it. Vapi is working on a native integration, which is going to be your best bet.
One of the things I love about your videos is the production quality. You are excellent in all you do sir.
I appreciate it!
Amazing breakdown as always Jannis!
Great Video! The community really needed this.
Thank you for all the value your provide!
Jannis is the ONLY way to go when it comes to Voice AI, MAKE and GHL
This is great! Do you know what I would do if I want to keep track of what user is calling? I'd want to somehow pass the userId from the UI down to a custom LLM where in the background it will use the userID to bring in the memory from previous interactions.
Thank you for another great video.
I really do want to see the details of how you customize the prompt. I have a vapi assistant with a script in the system prompt with branched logic depending on the user response. I would love to move that to a custom Ilm to be better able to control the flow, logic, and responses.
I have seen your “conversational pathways” video but it’s just not clicking for me.
A better understanding of how the python program and custom Ilm work together - like a flowchart.
You might want to look out for Vapi Workflows - will be released soon, and I believe that'll be super beneficial for your agents in general to have more granular control.
Great work ❤
can u integrate Agency Swarm into vapi and make a video showing how to do it? so that agency swarm can have complete control over the business and also do emails and sms etc. call backs and setup appointments ?
Another great video - love this stuff! I have to assume there's also a cost saving benefit even when going straight to Open AI. Right?
I don’t think Vapi charges extra for the OpenAI usage.
You might have more granular control of the length of the prompt.
For example, if the first message was said, you can change the prompt through a new one that removes the greeting part from it, which causes lower tokens during the call, which in return saves you a bit of money
@@jannismoore Yes! That makes sense...you're the Boss - thanks!
Thanks Jannis. Does using a custom LLM provide more security against using the VAPI provided models? Instead of VAPI taking the data from transcripts and it going to their model it is stored with ours?
It would certainly cut out one middleman. You can leverage a custom-hosted solution that is hosted based on your standards.
However, while it makes it a little more privacy-oriented, it's only as secure as your developers or devops are good.
Is the API key from openai a regular api key? Can you do another video on customLLM but using not only openai key but Assistant api key as well so then we can configure the assistant on openai and just use vapi for the voice. But instructions & functions are all done via openai! Also thank you for doing this video as a beginner! Literally the only thing I need help with is having vapi work using custom LLM connecting it to my assistant api (asst_key)
I’m working on that. :)
@@jannismoore Been trying to accomplish this also! Would love an explanation on this.. The endpoints for assistants are different to chat completions...
Thanks a bunch for another great video. Curious, I've used Air, and in Air all we need to provide is a call script. I'm now planning on learning Vapi, but I have no idea how to prompt in Vapi, to get an output similar to my current call script. Can you use call scripts in Vapi?
I think they use a custom Prompt to generate the master prompt including your transcript.
I'm not a fan of this as scripts usually have a lot of nuances.
It's best to custom prompt engineer them in Vapi for the best outcome.
@@jannismoore thanks. I was "afraid" of this :) I guess I'm better at call scripts than prompting. I'll give it a shot and see if I can achieve similar output through prompting.
Great video, thanks! Can you perhaps make a video on how using AnythingLLM (docker on a small VPS) as a custom llm for VAPI? They have an API endpoint compatible with OpenAI Completions. Somehow I can't get it to work.
Have you tried to push the code example into chatGPT with your question? That usually works pretty well
@@jannismoore I tried with chatGPT but didn't work unfortunately
How about using Hugging Face models. Would love a video on that
I think some can already be integrated with OpenRouter
Great video!! Do you have any lead generation video with Vapi?
Are you looking for more details on lead-gen with AI, or something like Outbound Caller templates?
Awesome, can this be made with laravel? Where can i find examples of it?
It can be done with Laravel, and I doubt there are any tutorials specifically on that available, but here's a prompt that let's you get started with ChatGPT:
Create a markdown formatted manual that generates a manual on how to connect Vapi with a Custom LLM Setup using Laravel. Ensure it's easy to understand.
Thanks for your great videos. Whenever I try to register to your Resource Centre, it won't let me register. Can you help please
Hey! Do you face any specific issue? There's a register link at the botttom - if there's something wrong with it, feel free to let me know
Thanks for your reply, I’m interested in outbound callers with Vapi, apologies for my confusion
Gotcha. I've got quite a few videos that are important when doing Outbound calling. Check out these:
- ruclips.net/video/lktoSWsHbiI/видео.html
- ruclips.net/video/3y7GJ9Vozkk/видео.html
- ruclips.net/video/iWZD_HBk9zQ/видео.html
Hey Jannis, i liked this video. im a beginner in the AI industry and i wanted to know if i can have a 5-10 min free call with you. is it possible?
Hey man, glad to see you join the space! I've limited my available timeslots, but you're most welcome to join my Discord. You'll find more about that on my channel page. :)
@@jannismoore Yeah thanks, ive joined your community. im hobby.
How does the server authentication work? I successfully made a POST to the /credential endpoint as per the docs, but I keep seeing a 401 error.
With the example I'm using there's no authentication as it's a proof of concept. The credential endpoint isn't even used. What exactly are you trying to achieve?
@@jannismoore I was referring to the authentication instructions from Vapi's docs on "Using your server" and I've got it working now, but the details in the docs are limited. For example, it doesn't say how to update the api key in the credentials or manage multiple environments.
HI, I want to use openai assistant api as my custom llm, can you tell me how can I do that
I can, but with the current setup, I don't recommend it as state management is incredibly difficult if it comes to things like function calling and interrupted messages. I've built it out before and had tons of issues with it. Vapi is working on a native integration, which is going to be your best bet.
i think vapi charges less when we use custom llm . Is that trueee
You basically pay the same, just with the difference that you pay OpenAl directly instead of doing that through Vapi
Bro, How much should i charge for an ai caller setup?
What do you offer?
@@jannismoore Sorry for late reply😅! Basically i setup appointment booking system to dentistry. What do you think, how much should i charge for it?
Where you from, bro??
Germany :)
@@jannismoore oh wonderful!
Can this be used for nfsw?
Yes
@@jannismoore only enterprise users?
@@jannismoore bro i just signed up on your info hub and your the best man!!!!! I'm truly appreciate that u made everything available for free!!!!!
@@Kevinsmithns No, anyone :)