The argument in one line.
Connecting Hermes AI agent to VAPI via MCP enables it to make autonomous outbound business calls, handle inbound customer inquiries, and run 24/7 cold-outreach campaigns—all orchestrated through plain English prompts for under pennies per call.
Read if. Skip if.
READ IF YOU ARE…
- A founder or freelancer with an existing service business who wants to automate appointment booking and lead follow-up without hiring a VA.
- Someone building AI agents or side projects who's comfortable with terminal commands and wants to see a production-ready phone integration example.
- A technical founder with 1-3 years of AI experience who's evaluated other agent frameworks and wants to understand Hermes' specific phone capabilities via VAPI.
SKIP IF…
- You're building consumer-facing products where you need production reliability, compliance, and call handling at scale — this covers hobbyist and small-business use cases only.
- You have no technical foundation and get stuck on terminal commands or API key setup — the walkthrough assumes comfort with CLIs, copy-pasting code, and basic troubleshooting.
The full version, fast.
Pairing an autonomous local agent like Hermes with a configurable voice platform like VAPI turns phone calls into another scriptable tool, available to you and to the agent on a schedule. The setup is one MCP install plus an API key: the agent then creates assistants, picks numbers, places calls, and reads logs without you touching the dashboard. Three patterns unlock most of the value: outbound errands like booking a massage at roughly five cents a call, scheduled cold outreach where a cron job pulls leads from a SQLite table the agent built itself, and inbound assistants that call back into the agent mid-conversation as a custom tool to fetch context. Pick GPT-5.1 over 4o, slow the voice, and add a voicemail-hangup rule before scaling.
Members feature
Chat with this breakdown.
Modern Creator members can chat with any breakdown — ask for the hook, quote a framework, find the exact transcript moment. Unlocks at T2: refer 3 friends + add your own API key.
Create a free account →Where the time goes.
00:00 – 01:00
01 · Hook + objection kill
Opens on whiteboard promise. Rapid-fires four objections (too complicated, too expensive, no business, not technical) and kills each in a single line.
01:00 – 04:00
02 · Install Hermes locally
One-liner installer, OpenRouter account setup, API key creation (names it 'subscribe' as a CTA), model selection (Opus 4.7), first test message.
04:00 – 07:00
03 · VAPI intro -- the missing piece
VAPI as phone-call configuration layer. Transcriber-LLM-voice pipeline explained. 10 free phone numbers per account. Sponsor disclosure.
07:00 – 10:00
04 · Connect VAPI MCP to Hermes
Three-step setup: get private API key, paste URL into Hermes, agent self-configures its own config file. No manual editing.
10:00 – 17:00
05 · Use case 1: Outbound booking call
Prompt: 'Research me spa massage in NYC today near Manhattan.' Hermes web-searches, picks Renew Day Spa, creates VAPI call. 19-second call, 5 cents. Post-call model and voice tuning.
17:00 – 24:00
06 · Use case 2: Cold outreach + cron jobs
Single prompt creates VAPI assistant Morgan from Brightlane with objection-handling system prompt. Places call -- 8-minute call hits voicemail. Second prompt builds SQLite DB and cron job for 24/7 automated outreach.
24:00 – 27:00
07 · Inbound voice agents
Hermes auto-creates inbound assistant on Polish number. Detects language, switches to English. 55 seconds from prompt to live deployed inbound agent.
27:00 – 31:01
08 · Use case 3: AskHermes as VAPI tool
Custom VAPI tool ask_hermes lets voice assistants call back into Hermes mid-conversation via ngrok. Live demo queries Hermes folder structure during a real call. Dual CTA: VAPI signup + free resource bundle.
Lines worth screenshotting.
- An AI agent with a real phone number can make outbound booking calls, chase leads, and operate on a cron schedule without any human trigger.
- A VAPI outbound call to book a spa appointment costs roughly 5 cents — the entire infrastructure for AI phone agents is now commodity-priced.
- Most power users already trust AI for email, coding, and browsing, but still make phone calls manually — that gap is where the next productivity unlock lives.
- One MCP installation is all it takes to give a local AI agent full phone capabilities through VAPI.
- An autonomous cold-outreach agent with SQLite call tracking can run 24/7 without supervision, logging every call result and skipping already-contacted businesses.
- Having a VAPI assistant call back into a Hermes agent mid-conversation as a custom tool creates a recursive AI loop that can handle complex real-time decisions.
- $5–$10 in API credits is enough to start running an AI phone agent — the barrier is not cost, it's configuration knowledge.
- Giving an AI agent persistent memory through SQLite means it never calls the same business twice and builds a cumulative outreach history across sessions.
Voice agents are not the future. They are already 5 cents a call.
Build this today
One MCP server turns your Claude Code or Hermes agent into a 24/7 phone operator -- bookings, cold outreach, inbound reception -- for less than a cup of coffee per hundred calls.
- The architecture: agent orchestrator + VAPI MCP = autonomous voice workflows. No custom code, no backend -- connect the MCP and prompt in plain English.
- VAPI gives you 10 free US phone numbers per account. Build an inbound agent before building any paid feature.
- The self-configuration pattern is the unlock: paste a URL into your agent and say 'set this up.' It writes its own config. Use this for any MCP you want to add.
- Cold outreach cron job is one sentence: 'call a new niche lead every 10 minutes, track in SQLite.' That is a complete outbound SDR for zero labor cost.
- The AskHermes-as-VAPI-tool pattern (use case 3) is the most underrated: a voice assistant calls back into your full agent toolkit mid-conversation via ngrok. Any live call can escalate to your entire agent stack.
Terms worth knowing.
- Hermes Agent
- An open-source, locally-installable AI agent framework that can autonomously complete tasks, use tools, and now — with VAPI integration — make and receive phone calls.
- VAPI
- A voice AI platform that provides phone numbers, voice synthesis, and call management infrastructure so developers can build AI agents that make and receive phone calls.
- MCP (Model Context Protocol)
- An open standard that lets AI agents connect to external tools and services through a unified interface, allowing capabilities like phone calls, database access, or web browsing to be added with a single install.
- Cron job
- A scheduled task on a computer or server that runs automatically at a specified time or interval — such as daily at 9am — without requiring a human to trigger it.
- SQLite
- A lightweight, file-based relational database that stores structured data locally without requiring a separate database server, often used for small-scale apps and agent memory.
- VPS (Virtual Private Server)
- A rented virtual machine hosted in a data center that runs continuously around the clock, used to host applications or agents that need to be always on without relying on a personal computer.
- Outbound call
- A phone call initiated by a person or system to contact someone else, as opposed to an inbound call where the other party calls in.
- Custom tool (VAPI)
- A webhook endpoint registered in VAPI that an AI voice agent can call mid-conversation to fetch external data, trigger actions, or consult another AI system.
Things they pointed at.
29:00productFree Phone Agent Swipe File (40 inbound + 40 outbound use cases + Vapi+Hermes map)
Lines you could clip.
05:30
“VAPI makes phone calls configurable while Hermes makes them autonomous.”
09:30
“You are probably underestimating these agents. You can literally say set this up and it will set it up.”
21:00
“The businesses that will adopt this will just crush the people who ignore it. It is that simple.”
26:20
“In fifty-five seconds, all of this is set up.”
Word for word.
00:00So I just gave Hermes agent a real phone number, and now it can pick up the phone to call businesses, book appointments, chase down leads, and even call you to confirm any details. But the wildest part is it can do so proactively on its own schedule without me having to ask it.
00:15This actually is the most powerful and useful Hermes setup that I don't see anyone covering. But, David, isn't this setup too complicated? No.
00:23It's literally one MCP installed and you're done. Okay. But what if it costs a fortune?
00:27It's actually super cheap. Alright? But I don't have a business.
00:30You will use it for personal things. Trust me. Dentist appointments, reservations, it can save you so much time.
00:35Or maybe you think you're not technical enough. You can't code. Listen.
00:38If you can copy paste, you can absolutely build this. So in this video, I'm gonna show you how to install Hermes agent locally, how to connect it to VAPI MCP, give Hermes a real phone number, then how to research target businesses, and have it call you a business, get more leads, and make you more money.
00:53Literally all done by Hermes agent. Alright. So the first step is to set up Hermes agent.
00:58I'm gonna show you how to do that locally. If you have a VPS, you can, of course, do it on a VPS. Either way, the setup is the same.
01:04So first, we go to the Hermes agent repo and we scroll down to find the one liner installer command. There it is. The quick install, just copy this command by clicking this button and open a terminal and then paste this in.
01:15Boom. This is gonna install Hermes agent on your computer. Once you run it, you should be able to type in Hermes setup and it will open up the setup.
01:21One of the first things it will ask you is what provider to use. So I'm gonna go with OpenRouter. Go to openrouter.ai, and then make sure to create an account.
01:29Super easy. Takes, like, twenty seconds. Then go to top right and click on credits.
01:33Make sure to charge up some amount. You don't need $300. Just 5 to $10 is enough.
01:38Go to left, click on API keys, and click on new key. I'm gonna name this one subscribe. If you're watching this, please subscribe.
01:45We are so close to 400,000 subscribers, and most of you, like 60% of you, are still not subscribed.
01:50So if you find my content valuable and if you want me to make more videos like this one, please go below the video and click the subscribe button. It's completely free. Okay.
01:59I'm gonna put some limit on this, maybe, like, $50, just a good precaution, and I'm gonna copy the API key. Let's switch back to the terminal.
02:05I'm gonna hit open router enter. I already have a key, so I'm gonna do r to replace it and paste in the new one.
02:13API key updated. Beautiful. We can choose a model.
02:16There's many different models. I'm gonna go with the Opus 4.7. It's asking to add a fallback credential for the same provider.
02:22I'm gonna do no. It's not needed for now. Text to speech, you can keep the current one.
02:26Hit enter. Enter. Iterations, enter.
02:29Enter. Enter. Just spam through this.
02:31The default settings are really good in Hermes. You can just click on enter a bunch of times. This is also good.
02:36Four AM, also good. Navigation platform, we actually don't need it, so I'm gonna skip by hitting enter. The CLI, that's cool.
02:44Well, I can just go to done, and we're probably done. And as it says here, always read the instructions, guys. A lot of people are scared of the terminal, and they don't look at the output.
02:52So I'm gonna type in Hermes, and this is gonna launch Hermes agent with Clothopus 4.7. Let's send a test message, test, And let's see if it responds.
03:02Okay. So we have Hermes agent installed locally. And, obviously, Hermes is a very, very powerful agent right off the box.
03:09However, there is one problem. It doesn't come with a phone number, and this is what we have to solve next. Real quick, if you want everything from this video, all the assets, all the repos, all the skills, everything I show you, click the second link below the video and grab it now.
03:22It's completely free. Okay. So we have Hermes, but the phone is the missing piece.
03:26This is still what most people don't trust the AI to do, to take phone calls for you. If you think about it, for 99% of power users, they already use AI for web browsing, for writing code, for doing their email. But using the phone, most of us still do it manually.
03:42Luckily, there's an elegant AI first solution, and that is VAPI. Basically, to explain it in the most simple way possible, VAPI makes phone calls configurable while her mess makes them autonomous.
03:53And the synergy between these two is really beautiful. Vapi gives Hermes phone numbers, inbound calls, outbound calls, voice agents, transcripts, call logs, and real time conversations. While Hermes gives Vapi goals, memory, tools, cron jobs, proactive decisions, outcome checking, next actions, self improvement, and anything else that RMS agent can do.
04:14So connecting these tools together is really where you get the most OP use cases possible. Let me show you how to set this up. Alright.
04:21So the first step is to just type in Wapi into Google and click on the first link. And here, simply create an account. So go to top right and click on get started.
04:28Now you can sign in with either Google, GitHub, or Discord or just classic email password. Now, again, creating an account is super simple. Literally takes twenty seconds.
04:36I already have account with Vappi, so I'm just gonna sign in. Now Vappi was kind enough to sponsor this video. So if you want to make your Hermes agent super powerful by giving them phone numbers and letting them do anything that a human with a phone can, make sure to show Vappi some love.
04:49Click the first link below the video and sign up to Vappy. First thing you're gonna see when you get here is the composer. This is this screen right here, and this is the agent that lets you build, debug, and manage your Vappy agents.
05:01And you can always access it from the left here, composer. You can even star it. So it's one of your favorites.
05:06So when you get to this page, go to left and click on assistance. This is where you can see all of the different voice agents or voice assistants you have inside of VAPI.
05:14This is the main page where you can manage all of the different voice assistants you have and see everything there is to know about them. Now the most interesting thing in here is the breakdown of the cost and latency for each assistant. And you can see that the average cost is just $0.1 per minute, which is very affordable.
05:30The average latency is just one point one five seconds or one thousand one hundred fifty milliseconds. Below it, you can see the transcriber, the model, and the voice voice model. Right?
05:39The voice agent. So, obviously, you can configure these and change these to different providers, different languages, a bunch of other settings.
05:45More on that later. But the main thing to understand is that this is the flow. Right?
05:48When a person speaks, the agent first needs to transcribe their speech into English or some text, then it sends to a model. Right? It could be a cheap model like 4.1, or you can use something like more powerful, like g b d five or even g b d 5.4 if you want even more power.
06:03And then the third thing is the voice. This is the output. Right?
06:06Once the model has the response, it needs to turn it into voice, into spoken sound that gets played to the person it's calling. Right?
06:13And, again, you can select the provider, the different voice preset, you know, male, female, slow, energetic, kinda more monotone. There's many different voices.
06:21You can adjust the speed background sounds. There is a lot of settings inside of VAPI. It's really, really advanced.
06:27And, again, I'm gonna go into this later in the video. Now before I show you more about the VAPI interface, let's make some progress, and let's actually set up Hermes so it can use VAPI. And we're gonna do that by using the VAPI MCP server.
06:40This is the simplest way to connect VAPI to Hermes so that it can actually start making calls on your behalf. And in fact, the team had made it super simple, and it's literally these three steps. Get the API key, edit the config, and add the MCP server block.
06:53So I'm gonna show you step by step. You can do this even if you're not a developer. And by the way, what you might not realize is that this will allow Hermes to create VAPI agents, to log the calls, to analyze what happened, basically, to make any changes inside of the VAPI dashboard for you.
07:07So, really, you don't even have to master the UI. All you need to do is talk to your Hermes in plain English, either in terminal or Discord, WhatsApp. I've made many videos on Hermes agents, so watch those.
07:18But, basically, all the heavy lifting will be done by Hermes. You just tell it what to do, and it can do it. It can interact with the full VAPI interface through the MCP.
07:26So really setting this up, this might be the most valuable three or four minutes of the entire video. So there's just one thing that you do. Please do this.
07:34Give your Hermes agent access to VAPI so it can make phone calls, it can set up these voice agents, and use them to save you time and make your life better and grow your business. I don't care how you use it. You have to use it.
07:45Okay? That's the main thing. Just make sure to use it.
07:47So first step, get the VAPI API clear. Literally click on this. By the way, I'm gonna link this, um, in the resources.
07:52Right? So, again, everything from this video will be linked in the second link blog video. Okay?
07:57All the skills, all the repos, everything you need is gonna be bundled together. Click the second link below the video. Grab it completely for free.
08:04Alright. So I'm gonna click on this link. It's gonna take me to the dashboard of VAPI, specifically the API key section, and I'm gonna create a new API key here.
08:12Now believe it or not, the simplest way to actually set this up is copy the URL, go to Hermes, and say, set up this new MCP. Boom. And let Hermes set it up for itself.
08:24Because Hermes comes with a skill that knows how it's architected. Right? It knows how it's built, and it knows how to set up MCP servers inside of itself.
08:32You are probably underestimating these agents, all of you. You're probably giving them too easy of a task, or you're micromanaging them too much. You can literally say, set this up and link what MCB server you want.
08:42It will set it up, and then it will ask you for the API key, which, again, that's the last step we have to do. So, again, if you click here, you'll get redirected to the VAPI dashboard, and you have the two keys.
08:52Now we need the private one here. But if you're not sure, just take a screenshot, go to Hermes, and ask it which of these keys do you need, and it will tell you. So there it is.
09:01Vapi, z, z, space. Okay. I need your API key.
09:03Again, just copy it right here. And by the way, anytime you're not sure which key, just screenshot, boom, copy to clipboard, paste it in with control v, and say, which key do you need?
09:14Answer in short. Private key server side. Okay?
09:17So it wants the private key. I'm gonna copy it. Say, here is the key.
09:21Store it securely. Obviously, this is not the best practice pasting API keys into the Hermes, and I'm gonna revoke my key before uploading this video.
09:29You can also do Hermes config set VAPI API key and then paste it in. That's the more secure way, but this is the fastest and the simplest way. And since Hermes will be basically doing everything else with the VAPI MCP, it's not that big of a risk.
09:41Alright. It's asking for permission to write this in. I'm gonna allow it for this session.
09:44And as you can see, it's setting every everything up. I haven't made a single change, single manual change to the config file of Hermes. It's setting it up itself.
09:53Like, this is the magic of self improving AI agents like Hermes. They can improve themselves. They can do all the things.
09:58K? It wants permissions for this command. I'm gonna allow for this session.
10:01And you basically just tell it what you want, and it will do it. Now there's two reasons why most people don't see the value in AI agents. Either they never set them up.
10:09Right, so they don't watch a video like this, or they only watch it and they don't follow the steps. And the second reason is they don't know what to tell it. They don't know how to ask it for what they want.
10:17So if you can actually set this up, if you can follow the steps I'm showing you and just do this yourself either on a VPS or locally, and then actually start telling it the things like, yo, make a phone call to that sauna. I need a reservation.
10:29Or find me 10 leads in this city and call the phone number on Google Maps. Like, if you can tell it what you need, it will do it. Issue is most people never set it up or don't tell it what to do.
10:39Okay. There it is. Confirmed.
10:41Here's the variable. Restart her mess to pick up the new server. Okay.
10:43So I'm gonna do control c to kill her mess. Clear. Launch it again.
10:48I'm gonna say check your config and let me know which MCP servers you see in there. Be concise. Just a quick double check to see that the VAPI MCP server is there set up correctly.
11:01There it is. Just one VAPI. Beautiful.
11:03Now check. Boom. And let me know if you see any API keys in there.
11:10Don't read the the values. Just let me know which keys you see. Okay.
11:16As you can see, already have some, but there it is. VAPI API key. Beautiful.
11:19It's there. So we can continue. Alright.
11:21So we have the API key here. We have the MCP server here. Before we jump to the first use case, I wanna show you a bit more about the UI in VAPI, especially the messages.
11:30Here, have the first message aka what happens. Right? Whether assistant speaks first or the assistant waits for the user or assistant speaks first with a model generated message.
11:39Right? So you can say, hey. I'm John or, I'm calling on behalf of David, stuff like that.
11:43And then the system prompts. This is the main thing. This is where you implement and influence the behavior of the voice assistant, how verbose it should be, whether it should be formal, informal, how it should talk, what level of English it should have, all of that.
11:56And then here, can select any external files, such as documents, PDF files, markdown files, anything else that would give it more context. But the setting we're looking for the most inside of VAPI is the phone numbers. If you go left, this is where we can see our phone numbers that we have.
12:11Right? So we I have two American ones, one Polish one, but this is what gives Hermes the ability to make phone calls through VAPI.
12:19Right? So these phone numbers, you can put different labels on it. You can see the provider.
12:23I think VAPI even gives you free phone numbers. If you click, uh, yeah, free phone number, you can get up to 10 per account. So if you're not convinced to try Vappi, this alone might be convincing.
12:3310 free US phone numbers for your AI assistants to make phone calls. Right?
12:38Very, very OP. But, yeah, as you can see, Vappi gives you all the configuration, and Hermes gives you the powerful AI agent can solve problems and do basically anything else, and now it can also take phone calls.
12:49So the first use case is gonna be having Hermes make an external call. So it could be book something or, you know, just call a business on your behalf, save you time, anything.
12:59Like, there's many situations where you might need someone to make a phone call. Right? So I'm gonna kill it.
13:04I'm gonna start her her mess from scratch, and I'm gonna give it a very simple prompt. Research me, spa massage in New York for today near Manhattan. Find their phone number.
13:14Okay? Very simple problem. Plain English.
13:16There's literally typos in there, but Hermes agent has all the tools and all the skills on how to do this, and we will do it right now. So does multiple web searches, best spa massage, here on top spa massages, boom, with the phone numbers.
13:28Very easy. Took a couple of seconds. It's already asking here, by the way.
13:31Want me to call any of these to confirm availability? Yes. Choose one of them and call them using VAPI right now.
13:39Calling renew day spa. Best walking availability using the appointment booking assistant with US phone number briefs her message today. Let's see this.
13:46This this is always exciting. We can actually go back to VAPI and probably see it in our assistance.
13:51So to use the appointment booking one, let's click here, and we can click on logs. Check status now.
13:57Let me reload here to see what's going on. Call in progress. Call connected.
14:00Conversation happening now. This crazy. I'm gonna confirm that it's the right assistant in VAPI.
14:06Which VAPI voice assistant is taking the call? Oh, wait. There it is.
14:09Look at this. We see it. Call ID, outbound duration, ninety seconds, cost 5¢.
14:15It cost us just 5¢ of VAPI credits to make this phone call. Let's look at it. Let's click on it and see what happened.
14:22We can see it right here. Okay? So we can even play it.
14:24Hi. I'm calling to check availability for a massage today.
14:28Today, what time? I'd like to know what times you have available. I wanted to a lot opening.
14:32You can tell me the price. Great. Thank you.
14:34Could you tell me what types of massages you offer and the pricing for a sixty minute session? Okay. So the call happened.
14:40Okay? Call happened
14:41was nineteen seconds. We need to probably fine tune the length of responses because here, it started speaking too soon, and then it interrupted the lady, and then the lady interrupted him.
14:53But this is a great start. Okay? As I said, in Vappi, you can fine tune everything.
14:58How fast the response, what model we're using, the tone of voice. We can change all of that, but the call happened. Cost us 5¢, which is incredible cost.
15:05I mean, think how much you would have to pay a human to sit on a call sit on a phone and do phone calls for you. Right? This is a huge first step.
15:13So, obviously, not every call is gonna be successful. This is still highly experimental technology. I mean, we're trying to be on the cutting edge here, but this is a great start, and we can actually fine tune a lot of this.
15:22When you click here, when you click on assistant, we can improve the model. G b d four o is very weak. I'm gonna go with something faster, maybe g b d 5.4 mini.
15:30I think that's a much better option. Or even g b d 5.4 itself, it's even better. Here, temperature controls randomness.
15:37I think, like, 0.7 is fine. All of this thing is good enough. Okay?
15:41So I think the model is setting. The Deepgram one, let's see. We have Deepgram, long language e English model Nova three.
15:49It's probably fine. Background denoising, numerals, profanity filter. K.
15:52All of this looks decent. Let's optimize the voice. Maybe we can try some of the VAPI voices.
15:57Let's see. Elliot. Maybe we can try a female voice like Emma from VAPI.
16:02Speed, I think a bit slower, maybe like one.
16:08Background sounds, let's try office and maybe for okay. For we don't need a full black voice.
16:13And then we can click publish, and it's gonna publish all these changes to the assistant. Looks good. Publish.
16:19And boom. Just like that, we've changed this assistant so that when Hermes uses it, it's gonna use a better AI model and a female voice model. Maybe this will be more successful.
16:27Obviously, just like with any anything in AI, you need to fine tune it. You need to play around with it, see what system prompt works best, and, yeah, test it out.
16:35But, honestly, this is a huge first win. The phone call happened. There were no errors, and it cost us just 5¢.
16:40But if you think this use case is crazy, just wait until you see what's coming. Because of the MCP, Hermes can create new assistants inside of VAPI optimized for any type of use case. So for example, I can say, create a new assistant inside of VAPI using the MCP.
16:57Make it about calls outreach to businesses to set appointments to get more leads. For example, let's focus on car detailing businesses in New Jersey.
17:06Create this assistant and do the first call. Plain English. Again, just prompt it in natural language and watch it do the task.
17:13It's initializing the agent. It will find the MCP. MCP VAPI list phone numbers.
17:18Boom. It checks the available phone numbers we have. K.
17:20It sees before it needs two things for me. Okay? Okay.
17:23Number one, you should research that yourself. And number two, just choose the first phone number. Get to work and actually do it.
17:28I feel like in this case, it didn't really need to ask me, but fair enough. So what we should do is inside of VAPI dashboard, inside of assistance, we should see a new assistant being created.
17:40There it is. MCP VAPI create assistant. Let's see what name it gives.
17:43Top tier Autospa. Hopefully, these businesses don't get too annoyed with our testing.
17:48K. We should have there it is. New Jersey car detailing legion outreach.
17:51Okay? Boom. Optimized agent with a custom system prompt.
17:56You are Morgan, SDR from Brightlane, Legioneration Agency. Look at this. Hermes agent wrote all of this.
18:02Objection handling. Damn. Be warm, casual, confident, American English.
18:07This is not bad at all. Okay. MCPOP create call.
18:13It's doing a call. Wait. Wait.
18:15Wait. We should see logs for this guy. No session.
18:18Okay. Nothing yet. Call started.
18:22Check progress on this call. This is amazing.
18:25Like, the synergy between Hermes and VaPi is insane. Call failed, never connected. Okay.
18:30Yes. Try the other number. Last time, it didn't have this issue, so I don't know what's what's the problem here.
18:38Cool. Make sure to auto check-in thirty seconds.
18:43And by the way, you can run this in parallel. So you can have multiple different voice assistants on Vappy, all orchestrated with Hermes to run multiple calls in parallel.
18:53So one of them could be set booking appointments, one of them could be confirming calls, one of them could be qualifying leads. Like, whatever you can imagine, you can build. It really is up to you to figure out the use cases.
19:05And I'm showing you just a couple to get inspired, but really sit aside and think about this. Because the businesses that will adopt this will just crush the people who ignore it.
19:15It's that simple. The world really is turning into people who embrace AI and people who hate AI. There is like, it's a it's a barbell.
19:23Right? Both of these ends, there's nothing in the middle. It's just people who embrace AI and they love it because they can build any app.
19:29They're way more productive. They can do things like this, and then people who hate AI. It's more more like, I guess, left leaning, but college students.
19:36This is a big, like, you know, artists. There's a big hate towards AI, and those people are just gonna get crushed.
19:42I mean, it is what it is. You know? Either they adopt a new technology or they're gonna be left behind.
19:47I mean, you and me know that. We're you know, you're watching this video, so hopefully, you're on the future side of the AI equation. But I'm telling you, this is the landscape.
19:55It's either people who fully embrace AI and people who hate it. There's nothing in between. Live now in progress.
20:00Okay. It's happening.
20:05I don't know if it's a good talk because I'm in Poland in Katowice. If you wanna join our Katowice office, click the link below. We're hiring for multiple positions.
20:12But this is this is good. We're calling American companies, so it's probably morning in America.
20:18The status there, call is still alive. Okay. Check again.
20:22I wanna see what's happening. This is exciting. Anytime agent is on the call, this is super exciting.
20:27And by the way, another thing is the advantage of Hermes is that you can create a cron job. So you can do these calls automatically. So I could say, create a cron job that every ten minutes calls a different lead from New Jersey that is a car detailing business.
20:45Build a SQLite DB to keep track which businesses you've already called.
20:54And literally with a single prompt, we have a outreach strategy. We have a cold outreach strategy to slowly and surely well, actually, not slowly, quickly, like every 10 minutes.
21:03Right? And it's gonna work twenty four seven, so even humans cannot work twenty four seven, to call every car detailing company from New Jersey.
21:10Right? And it's gonna build SQLite database to keep track. Okay.
21:13Did I follow-up with these guys? Did I already call with these guys? And you can use that context in the future calls.
21:17Like, people don't realize how easy it is to build something that's truly useful, something that can grow your business today, not hypothetically in the future.
21:28If you need car detailing leads or if you need chiropractors, gyms, anything, like, anything.
21:34This is not just from, you know, brick and mortar businesses. Like, get creative.
21:38You need to call doctors. You need to call lawyers. You need to call finance firms, private equity firms, whatever.
21:44Right? It can do these phone calls. You can fine tune the the system prompt, and then it just say, Hermes.
21:51Okay. Clear new crunch up that every fifteen minutes, it rings a different venture capital firm pitching my start up.
21:57Boom. Would you pitch your start up faster? Would you get investors faster if you build this?
22:02Probably, It's like, get creative with this, guys. Use it for your business.
22:06Use it for something. This is truly, like it's truly incredible what you can do.
22:10Anyways, in the meantime, Hermes is creating Cronjob to build our automated outreach system. And, I just said a single sentence. Single sentence on what it should do, and it created SQL database.
22:21It's it's set up the MCP. It's created the skill, cron job, all of that because Hermes counts with, like, 82 prebuilt skills. So it it knows how to do this.
22:30Okay. I can actually do another Hermes, and I can copy this assistant, use the VAPI MCP to check on this voice assistant and progress on most recent call.
22:43Probably could have given it just the call ID to be fair. It's probably better. Here is the call ID for Vappi.
22:52So we're running multiple Hermes in parallel, which you can do very easily. Oh, there it is.
22:58The reason customer wait. Wait. Wait.
23:00This is insane. Eight minute call. Yo, guys.
23:03This is crazy. Eight minute call. Okay.
23:05Is this voice mail?
23:06Please, no. Lead to the following menu options.
23:09For operating hours, press 1. For address Okay. So it got stuck on voice mail.
23:14Alright. That's that's unlucky. Okay?
23:16So here what you should do is you should go to assistant and update the rules.
23:22Right? If you notice voice mail, just hang up.
23:27Boom. Something like this. Then I would go to model.
23:31This is definitely something we should change. Four o cluster is not good enough. 5.4 would recognize this much better.
23:38So, obviously, inside of her mess, it created the skill. Say, I'm gonna say update the skill you created for VAPI so that you always use g p t 5.4 as the model instead of g p t four o.
23:50That was a mistake from Hermes. It shows a worse model. So, yeah, I'm gonna publish this change so that we don't get stuck on voice mail in the future.
23:58Now before I show you the second VAPI use case, which is inbound agents, right, when you get inbound calls, let me first explain how VAPI actually works. So VAPI is the system and the platform used to build AI voice agents. Basically, apps that can talk to people over phone calls in real time.
24:13Now the most important elements of VAPI are the transcriber, the large language model, the voice model, and the orchestrator. And as you might have noticed, VAPI gives you all of this in a single platform. Now the orchestrator, this might be the most important one.
24:24It stops when people interrupt. It cuts our background noise. It knows when you're done talking and picks up on the mood of the people talking to your agent.
24:31Right? So all of these are very those are the hard part. There is voice models.
24:34There is LLMs from every company. But putting it all together to something that understands these nuances of phone calls, that is the hard part. Not just having generated AI voice, but making it into a truly useful assistant.
24:48That is what Vappi did. Now to show you the second use case, you can just create Hermes. And let's say you run a business that gets a lot of inbound calls.
24:56You can say, I run a massage place in Katowice, Poland, and we get a lot of inbound calls.
25:03I want you to use the VAPI MCP to create a new voice assistant for handling these inbound calls to just let them know about our availability, you know, which hours and which type of massages we offer, and help them book and schedule stuff using our internal software.
25:21I'm gonna give you more context later, but just use the VAPI MCP to create this new inbound voice assistant. If you have a business like that, you probably are paying two, three, four people on retainer every single month to just handle the calls. Right?
25:36To to handle the, you know, like, secretary or to for the phone call. Or you just deploy one VAPI assistant to do that twenty four seven, way cheaper. And if you put an effort into the system prompt, more reliably because, again, you can just customize this.
25:49You can test out different transcribers, different LLMs, different voice models, test out different opening messages, system prompts, connect it with other tools.
25:57Right? So if you have your own internal software, your own API, you can connect this here. Uh, as you can see, there's many different tools here we have.
26:04Prebuild, we can also create a new tool and put in the schema and connect it to your software or maybe Zapier or n a 10, whatever use. There is a lot of possibilities.
26:13So if you have a business that takes inbound, I would say it's nonnegotiable. You absolute absolutely have to do this. In okay.
26:19This interesting Hermes realized that we have the Polish number. It assumed that's for that, and it created a new assistant. Language Polish auto switches to English.
26:27This is crazy. In fifty five seconds, all of this is set up. In fifty five seconds, I can deploy this with inbound.
26:33Voice agents are here, and I showed you two use cases. Let me show you the third one. The third use case is the most impressive by far, and it uses custom tools.
26:41So inside of Vapi, go to left, click on tools. And here you can see a bunch of tools. A lot of them are prebuilt.
26:46But one is custom, ask Hermes. Right? And this is the way you can use Hermes agent as your personal concierge.
26:52So instead of Hermes delegating to Vapi, Vapi can contact Hermes. So during the conversation, let's say, in a client, right, with a client, call with a client and it needs more info about your business or something, it can contact your personal Hermes agent deployed on the VPS and ask him questions and get contacts with him.
27:10Maybe the Hermes agent can send you a DM in the middle of the call on Discord or Telegram or WhatsApp, wherever your Hermes is authenticated. They're like, yo.
27:19The the this client is very promising. He wants to talk to you directly. Like, stuff like that.
27:24Right? It really is insane what you can do with this, and I have this tool here, AskRMS. It uses Angular to create a tunnel, and this really takes it to the next level.
27:33As I told you, VAPI is very, very configurable. So let's test this out. I'm gonna copy this ID of the tool, switch back to the terminal.
27:40I'm gonna do control c to just kill this to create a fresh hermes session. I know I can do slash new, but this is cleaner. I'll say list out all tools, all MCPs we have with VAPI.
27:53Then use this tool to ask Hermes something specific, perhaps to list out its internal folder structure.
28:05VAPI gets tool. Okay. Hopefully, it gets the but also get okay.
28:10Tools. Boom. Here's the ID.
28:12And this really fills in the last gap where if the VAPI assistant isn't sure on what to do during a call, it can use AskHermes to ask your Hermes agent, which can have connections to a lot more things and, you know, usually uses a more expensive model and has more context about you and the business to complete the request.
28:29So inside of the VAPI dashboard, I have this assistance, and I have this Hermes voice concierge. Here, have the tool attached.
28:36Ask Hermes. You need to give the specific tool to that specific assistant so it can use it, and we can actually talk to it. Right?
28:43So we can click on talk and do a test. I need to allow the microphone.
28:46Click allow. Hey. This is your AI assistant.
28:49What can I help you with?
28:51Hey.
28:53What tools do you have available? You have quite a few quite a few tools, browser control, terminal and file editing on your VPS, web search, VAPI management, cron jobs, sending messages, memory, skills, and delegating to sub agents.
29:08Want me to explain any of these in more detail?
29:10I want you to ask Hermes and tell it what is the internal file structure and folder structure. That's really good, guys. We got the ask Hermes tool here.
29:19It obviously takes a few seconds because it hits a VPS server. There it is. Hit again.
29:22Top level has Hermes and WAF root on Hermes, your VAPI adapter, and opt VAPI adapter,
29:28plus the usual root home dot files. Want me to dive deeper into the Hermes folder or the adapter?
29:34Really nice. So you can attach this as a tool to any of the voice assistants, and it can use it during a conversation.
29:41Right? So now I'm testing it myself, but, obviously, ideally, it's a potential client, a potential lead, maybe an applicant, whatever.
29:46There is, like, hundreds of use cases for voice agents. But now you know how with VAPI and the tools feature, you can use Hermes not only as a orchestrator, because so far, we've used Hermes to manage and create VAPI assistants. Right?
30:01But now you can use you can do the other way, where, basically, a VAPI assistant during the conversation can call to Hermes to get, you know, more more context, more tools, more connections, whatever is needed.
30:14So this really is the ultimate platform. With VAPI, you can build anything you imagine when it comes to voice agents, especially when you connect with with Hermes. So, again, I have to thank VAPI for sponsoring this video.
30:24And if you have a business, whether it's one that gets inbound or one that could use outbound, please start building voice agents.
30:31I don't care how you use it. You just need to start using it. Even if it takes you a couple weeks to figure out and really crack the code, get started now.
30:37Click the first link below the video. Give Vappi a shot. They have multiple different plans, so choose the one that makes sense for you and actually build this today.
30:44And as I mentioned before, if you want all of the resources from this video, for the different use cases for inbound, for the different use cases for outbound, and everything else mentioned in this video, click the second link below the video. It's completely free. You'll get a bundle with everything I mentioned in this video.
30:58So go below, click the second link, and grab it now.
The Hook
The bait, then the rug-pull.
A phone number. That is the one thing separating a capable AI agent from an autonomous one. David Ondrej spent 31 minutes proving that the gap is exactly one MCP server wide -- and by the end, his Hermes agent had booked a massage, cold-called car detailing shops in New Jersey, and answered his Polish phone line in English, all without him touching the keyboard.
Named ideas worth stealing.
05:30model
VAPI + Hermes Synergy Stack
- VAPI gives Hermes: phone numbers, inbound/outbound calls, voice agents, transcripts, call logs
- Hermes gives VAPI: goals, memory, tools, cron jobs, proactive decisions, outcome checking, self-improvement
Core mental model: VAPI handles phone-call infrastructure, Hermes handles autonomous decision-making. Together they handle any voice workflow.
Steal forAny pitch for MCN+, JoeFlow, or agent-stack products that combine a low-level tool with an AI orchestrator
09:20concept
Agent Self-Configuration Pattern
Give the agent a URL and say 'set this up.' The agent reads its own architecture docs, writes its own config, and confirms. Works because Hermes ships with 82 prebuilt skills including self-setup.
Steal forPositioning any AI tool that removes manual configuration steps
10:00list
Three Voice Agent Use Cases (escalating complexity)
- Outbound booking -- research + call a specific business on your behalf
- Cold outreach + cron -- create SDR agent, build lead DB, run 24/7 calling cadence
- Inbound reception -- answer your phone number, qualify leads, route calls
Consumer to business to enterprise. Each demo is live with real cost and call logs visible.
Steal forContent structure for any 'X use cases for Y' tutorial format
How they asked for the click.
30:00link
“Click the first link below the video. Give VAPI a shot. And click the second link -- completely free. You will get a bundle with everything I mentioned in this video.”
Double CTA: sponsored VAPI link + free resource bundle. Free bundle framed as standalone value. Both pushed multiple times throughout the video.
Visual structure at a glance.
