vocode

An open-source platform for building, deploying, and scaling hyper-realistic voice AI agents across various applications.

Vocode

Hello who am I talking to? Hi where can I learn more about Janis? Hi there you can learn more about Yanis Moore on his social media channels. He is active on YouTube, Instagram, Tik Tok, and Twitter. He shares Ines about Automation and educates his community there. 

That's awesome thank you very much. You're welcome if nice. So the past two months have been an immense learning curve for me, especially regarding AI and voice. And now one month after my video on how to create an AI voice agent using sflow, many many other YouTubers saw the same potential in voice AI and started creating videos as well. 

So while they have been busy understanding sflow, I've had the pleasure to deep dive into other tools that are still pretty much unknown. This video is your ticket into a tool that I discovered during my research. It is called Vocode and it is essentially an open source framework that you can use to build custom voice AI agents on top of it without taking the route through an additional provider. 

Like this allows you to have phone calls for less than 2 cents per minute while being able to completely customize everything without limits. While wot anent developers, I will show you how you can make it without writing any lines of code and without any knowledge about coding whatsoever. And of course with as little cost as possible. 

Vocode also offers a hosted solution in case you do not want to host the backend yourself, but I found it very limiting in the set of features they offer through the API. To make everything work we need a couple of SAS tools. So in case you don't have an account with them already, please go ahead and create one. 

Number one is going to be open AI API. So open AI is basically the LLM we are going to use on which our AI agent is powered. The second tool we are going to use is called Deep grum and it is used to convert text to speech and speech to text. 

Number three is called render which is the platform on which we host our very own AI agent. In our case we are going to use a free plan so you can too. Number four is GitHub so that you can manage and extend your AI agent. And lastly we are going to use Twilio as.

Our phone number provider so that we actually able to call our AI agent as you have seen in the beginning. So with that said let's get right into it. The first thing you need to do is creating a GitHub account. Once you did that in your logged in you can head over to my AI voice agent template that I created for you. This one basically contains everything you need to create that agent without writing a single line of code which is in fact a template that I kind of created before and try to use to create my own assistant with which also contains transcript which also contains an integration with 11 laps in case you want to use it. 

In this example we are not going to use it to just reduce the amount of tools we need but obviously if you want to use custom voices you can do that as well. Lastly I packaged all of that in a Docker container so that we can simply deploy it with render.png coming videos. Once you're logged in and you are at our or at my repository you need to head to the fork button right here and fork the template into your very own account. Since I'm the owner of this template or of this repository I cannot do that but you will definitely be able to do that. It's also free so you simply click on fork you fork the template it's inside of your account and once you did that you simply go ahead and copy that URL in case you already have tons of repositories. 

If not you probably don't need to worry about that cuz for now you basically set up everything inside of GitHub. The next step you need to do is head over and create an account on render.png your GitHub account. If you haven't done that yet you can do that only with the repository you want or with all of them. This is really up to you. I would always suggest to just select the repository you want render to have access to for security reasons. Once you've done that you simply click on next and you will see a list of repositories. Once you're here you simply locate the AI voice agent voco template which in my case is that one and all we are going to do is click on connect which basically brings you to the next page where we can configure that whole.

Template mine basically adds it with A-1 as I already have one deployed. I'm simply going to show you how you can set it up on render and then I will show you how it looks like if you have it live. So the name basically it's not really relevant in our case. You can simply set it to whatever you want. I would always justest s bit standard. 

The region is a bit more interesting because it kind of like depends on the way you want to host that AI agent. I would always suggest to choose something that is closest to you or closest to the customers you're going to serve. In my case, I'm simply going to use Frankfurt cuz it's anyways I'm not going to set it up. But if you let's say are from the US, you can use Ohio or Oregon or if you're from Asia it's probably better to use the one in Singapore. 

And once you've done that, you simply go down. You leave the Branch as it is. You don't need to change anything on the root directory and you also don't need to change anything for the run time. All of that is basically pre-selected and preset by render in a way that it will work. So you literally can ignore all of that. 

And if you are like me and you just want to test the bot stuff, you can simply select the free plan as we don't need to actively have it um available. What I mean by that is that the free version basically always spins down after a time of inactivity. So in case no one is using the service, it will automatically spin down and then when you try to access it, you need to wait a while until everything is pulled up again, which in that case can take a couple of minutes. 

Once you've selected the free plan, you can scroll down and you will see the last part where we take care of and this is the environment variables. To fill out that one, we need a little bit of time because the first thing you need to do is head over back to the GitHub template that you just forked. You scroll down until you see the read me file right here and you look at 0 five which is the environment variables. 

So in case you don't want to follow the video, you can also simply go through the manual and basically set up that stuff there. But for us since I'm

Going to explain it to you. I'm simply going to start with the environment variables and the first one is obviously to set up and connect OpenAI. So all you do is you simply copy the OpenAI API key environment variable. You add it here and you would add your OpenAI key here. You will find it by heading over to OpenAI. You log into your dashboard, you go to the API key tab, and you copy the API key. So that is literally one of the simplest steps.

Then we head back into our GitHub template and we are continuing with the, let's say, Deepgram API key. I'm just going to do it like that 'cause it's easier for me to set up and better for you to understand and follow along. So Deepgram is basically the tool, like I mentioned, that we are going to use for speech to text and for text to speech within our call for streaming. All we're going to do is we head back to Render, we click on add environment variable, we add it here, and we would add our Deepgram API key here. Deepgram is that platform. All you need to do is create an account or sign into yours. We have $200 worth of credits in the beginning. That is more than enough for you to test the bot for a while, for a long time, and you will definitely also get access to an API key that you then simply copy and paste right within this field.

After that, we are going to head back into here and we are setting up Twilio. So in case you haven't created a Twilio account yet, please go ahead and do that. Within the account dashboard, you will find the Twilio account SID and the Twilio account token. So you would basically simply add two environment variables here, first this one and then the one for the auth token, and you would add both of them right within here. So I basically have done that already with my other server, so I'm just using demo data right here. 

And there is not much more to that as of now except for the transcript callback URL, which I think is like the most important one in case you actually want to develop an API with that. That makes a lot of sense and that you can use in real examples 'cause this callback URL is used whenever someone calls or interacts with your AI agent. We basically take the transcript.

And send it to the web URL so that you can access it from wherever. Right, so in my example I basically did that before in the test that you saw earlier. I basically just asked that and it basically sends it to a URL. So this site is called web hook.

Site which allows you to test those kind of web hook endpoints. So in case you would set up something like webook to side for testing, all you would need to do is you click add environment variable. You add the transcriptor call URL. You go to web hook.

Site and you copy the web hook URL that you will see right here. So this is basically the one we can send the information to. So we copy it and we will paste it here, which means that every time someone calls and finishes a call, that information or the transcript will be sent to this web URL, which then looks something like this in the example that I created for you. I'm not going to go into the details. I'm just keeping that for now to keep the tutorial as slim as possible. If you want to read something about it, I definitely suggest to look into the repository. Everything is documented there. 

But for now, let's just continue with the setup because once you have done that, all you need to do is you need to click on create web service, which is what then deploys the service on a server so that you can use it live. So I have already done that for you, so I'm simply heading over quickly to the server that I set up. So once within the server that you created, it looks something like this and you will see within the events the deployment, which basically means that your chatbot has been deployed from the repository inside of GitHub. 

And if that works, which should out of the box, you should be able to use the chatbot right away. Not right away, but nearly right away, 'cause there's one more thing we need to do, and that is basically connecting the Twilio phone number to our AI agent. So to do that, the first thing you need to do is you need to head over to Twilio. Once you are inside of Twilio, you would need to purchase a phone number. To do that, you simply go to phone number manage active numbers and you buy a number, or you simply click the button up here. I think it's also available in the main page. 

And once you have an active number, it will look something like this, as you can see here. In my case, we basically need to set up this inbound call URL. This is basically what you need to set up to make the whole thing work. Before I'm going to do that, I also want to mention that in case you have a trial account with Twilio, I suggest to upgrade to an actual paid one or to a real account 'cause the trial account has sometimes some issues.

With voice calling I've tried it before. It didn't work all the time or it was a little bit buggy. So definitely try to just set up an account that is already out of the trial. That will definitely help you in the sense of testing the chatboard. The number price depends on what number you choose, but I would say it's usually between $1 or $2. 

So once you have your number here, all you do is you click on it. You wait until the page loads. You scroll down here and you look for The Voice conf configuration. This is basically where you can see a car comes in. It's set to webook and it sets a URL with the HTTP type, the request type post. So you want to set this part up nearly the exact same. The only difference is that you basically have to copy the URL of your deployed server and on render. 

So once you set up your server and render successfully, it is deployed. You will also see a URL right here. All you do is you copy that to the clipboard. You head into twio and you add that URL right in here. At the very end, you add a slash and write inbound on _ call. This is literally all you need to do. After, you can simply save the active phone number and you are ready to go. 

This is really all you need to do to set up that phone agent. This is one that is completely custom hosted on a server on render that is not controlled by anyone like sflow. So you can scale it up the way you need it. The best part of it is you can even extend it to whatever functionality you need. This also includes custom actions. It includes the whole settings, literally the whole package. 

If you now want to test the AI agent, you can simply use the number that you created and call it. You will then be able to see if it works inside of render in the locks. If by clicking here and within the locks, you will be able to see the whole transcript or basically like everything that happens. Right? So if I scroll up here a little bit, you will be able to see stuff like this where it says that's awesome thank you very much. That's basically what I said in the beginning of the example or in my first phone call that you saw in the very beginning of the video.

This is literally it. So you can look through everything here. You can also check after the phone call if the web hook was sent properly to web hooked side and if you see everything. 

So now since you have set up the chatbot, you might think yourself, "But Jiannis, I'm not a developer. How can I make it work?" And the answer is simple. It's called outsourcing. The easiest way of getting it done is by hiring a Python expert either on Upwork, which is my preferred source, or on Fiverr. 

To make the process as smooth as possible for you, I've created for you an example job description that you can use on Upwork. You literally can just copy and paste the thing to find a developer that specifically fits your needs. You will find the job description with all the other details and everything else I showed you in this video inside of our resource hub. So simply head over there. The link for that is down below in the description, and you can just get started with the whole thing. 

I have used the same process for one of my projects, and the developer managed to finalize the project in less than one week. This one job led me to 175 applicants, which gave me more than enough options to choose from. This is specifically regarding the developers I found on Upwork. 

Once you hire a developer, you can simply get him access to the GitHub repository that I created or that you forked earlier. So to get your developer access to your GitHub repository that you forked from my template, all you need to do is head over to it. You go to settings, you go to collaborators, and you add the collaborator right within there. 

This allows the developer to commit their very own changes so that you can simply look over what he did and communicate with him in case you want any changes or obviously you're not happy with his quality of code, which I hope will not happen if you're using the prompt that I provided to you within our resource hub. 

With that information, you now gain access to another neat platform of building completely custom voice AI solutions, and those are mostly cheaper than all of the other providers that I'm going to show you in the future.

Videos and not only that but you will also have full control over them. So it's definitely worth looking into. For now I hope you enjoyed this video. 

And definitely make sure to share AI Agent Store with colleagues in case you find this information helpful. As you are one of the few that made it to the very end, I'm going to share a little secret with you. There are a lot of more exciting AI tools on the market that haven't really been recognized. 

I'm going to change that in the next video. So definitely follow until then. Have a good one and see you next time.

Note, that there may be slight differences in text and video. Watch video for the original content.

Vocode Video