[0:00] This tiny computer runs my Open Claw
[0:02] locally 24 hours a day. No cloud APIs,
[0:06] no token costs, and even if the internet
[0:09] goes down, it's still working. But
[0:10] getting this system working was way
[0:12] harder than I thought. I had to test
[0:14] different local models, configure
[0:16] networking, and even split the system on
[0:19] two machines. In this video, I'm going
[0:20] to show you how to run Local Claw
[0:22] locally, how to pick the right models
[0:24] for your computer, and a setup I'm using
[0:26] at home with a Jetson Nano and an old
[0:29] gaming laptop. Why should you run Open
[0:31] Claw locally? First of all, Open Claw is
[0:34] expensive to run. It eats up a lot of
[0:35] tokens and before you know it, you run
[0:38] out of credits or you're spending $100,
[0:40] $200 on your Open Claw. Secondly, if
[0:43] you're concerned with privacy, you don't
[0:44] want to send all your data to a public
[0:46] LLM.
[0:48] Everything stays within your network and
[0:50] nobody else can see your data. Most
[0:52] importantly, what I found was that my
[0:55] open claw keeps going down because
[0:57] either I'm out of credits or claw just
[1:00] went down the other day. The servers all
[1:02] went down and I couldn't call anymore.
[1:05] So when your cloud provider is not
[1:07] working for whatever reason, it's
[1:08] updated as model. You always have
[1:10] something running as long as your local
[1:12] server is running. So your open claw is
[1:14] available to you at all times. Another
[1:16] thing that caught me by surprise is that
[1:18] when I first started using Open Claw, I
[1:20] was able to use it with my OpenAI
[1:22] subscription, my Claude code
[1:24] subscription, and my Gemini
[1:26] subscription. And now Gemini and Claude
[1:28] bans users who use the pro plan with
[1:31] their open claw. So now you don't have
[1:33] to worry about the policies of all the
[1:36] different AI providers. You have full
[1:38] control. But I must also say that the
[1:41] concept of local AI is really good. But
[1:44] getting it running locally, it's really
[1:46] hard. Complex setup, you have to
[1:48] understand networking configurations.
[1:51] You have to have the hardware. You need
[1:53] to have a fast computer. Otherwise, it's
[1:55] just super slow. And it's taken me more
[1:58] time to fix and configure things than I
[2:00] would like. Running Open Claw locally
[2:03] gives you a lot of freedom, but you also
[2:05] become the system administrator. But
[2:07] what does running open claw locally
[2:09] mean? Actually, I think there's two
[2:11] components. Number one is where does
[2:13] your open claw run? And then open claw
[2:16] needs to call an AI model to process the
[2:20] request. And where is that AI model
[2:22] running? You can have a cloud setup
[2:24] where your open claw is hosted on a
[2:26] server somewhere on Amazon on hosting
[2:29] wherever it is and then that open claw
[2:32] calls open AAI or claude. That's a fully
[2:36] cloud setup or you can have a fully
[2:38] local setup where you have a machine and
[2:40] it runs open claw and on the machine
[2:43] have a local llm model running so it can
[2:46] provide the responses to your open claw
[2:48] or you can have a hybrid setup where you
[2:50] buy a Mac mini and it's running openclaw
[2:53] right in front of you in your house but
[2:55] it's calling open AI cloud gemini and
[2:58] that openclaw calls a cloud LLM. So,
[3:02] we're going to cover how to run OpenClaw
[3:05] locally on your own device and hosting
[3:08] the AI model locally within your house.
[3:11] We're going to go through two examples
[3:13] in this video. The first one is the
[3:14] beginner setup, which is everything on
[3:16] one machine. I'm going to show you how
[3:18] to install OAMA and then using OAMA,
[3:22] install Open Claw, and then run a local
[3:25] model. So, that's the beginner setup and
[3:27] then we're going to go [clears throat]
[3:27] into my current setup. I have a Jetson
[3:30] Nano which is like a tiny computer
[3:32] almost like a Raspberry Pi running open
[3:34] claw and then I have an old gaming
[3:37] laptop that's running Lama and serving
[3:39] the AI model. Why? Because number one, I
[3:42] don't want to run open claw on my
[3:44] MacBook because of security reasons. And
[3:46] number two, my MacBook is really slow.
[3:49] So when I talk to Open Claw, it takes a
[3:51] long time for it to respond. So, by
[3:53] running it on a old gaming laptop, I get
[3:57] much better performance. And I'm going
[3:59] to go into how to set that up. And I
[4:00] also want to show how you don't have to
[4:03] buy a super powerful computer like a Mac
[4:06] Mini to do this. You can do this on some
[4:10] computer you have lying around, an old
[4:12] gaming laptop, and put them together and
[4:14] make your own local Open Claw setup.
[4:17] Okay, enough talking. Now, let me show
[4:18] you step by step how to set up Open Claw
[4:21] on your own computer. The first thing
[4:23] we're going to do is go to alama.com and
[4:25] install Alama. There are two ways to do
[4:27] it. Either you can run this terminal
[4:29] command or you can download a lama. So
[4:32] you can press download and install it
[4:33] there. But the best way to use a lama is
[4:36] through the terminal. I'm going to copy
[4:37] this command and then just run the
[4:39] command. And then I can run a lama by
[4:41] typing in a lama. And now I can run a
[4:44] model launch cloud code launch codeex
[4:48] launch open claw. So, let's start with
[4:51] running a model. And then you're allowed
[4:54] to choose different models. And it's
[4:56] giving me recommendations based on my
[4:58] specs. The recommended list is not the
[5:00] best. So, you can choose, you know, GLM
[5:04] 4.7 Flash if you wish just to get
[5:06] started. I already have a model
[5:07] downloader, so I'm going to use that.
[5:09] And I'm going to show you how to pick
[5:11] the best one and update that later. And
[5:13] I'm going to give it a test. Hi. All
[5:14] right. And so, Alama is running. And now
[5:17] let's download Quen 3.5. So I go to a
[5:20] llama. I click on models. I click on
[5:23] Quen 3.5 and I'm going to select Quen
[5:26] 3.59B latest. Copy this. And then I type
[5:30] in OAMA run and then paste Quen 3.59B.
[5:35] And it's going to start downloading the
[5:37] model. And let me run that model. Okay.
[5:39] And it's done. So let's give it a test.
[5:40] Hi. And it works. The next thing we're
[5:43] going to do is a new thing that open
[5:46] claw has enabled which is you can now
[5:49] use lama to install open claw. To
[5:51] install open claw with lama all you need
[5:54] to do is copy this command. Alama launch
[5:57] open claw. Copy it. Then go to your
[5:59] terminal and paste that in. And then
[6:01] it's telling me to choose my model. I'm
[6:03] going to choose quen 3.59b. I understand
[6:05] the risk. Okay. So, it's finished
[6:07] installing and I just sent it a message
[6:10] saying, you know, I'm Keith and it
[6:11] responded, but it's taking a very long
[6:14] time. So, the problem with Quen 3.59B is
[6:18] that it's got reasoning and it thinks a
[6:20] lot before it does that. So, I'm going
[6:22] to tell Open Claw to set it to no think
[6:24] mode. It's taking too long to respond.
[6:27] Okay, now it's set it to no thinking
[6:30] mode. So, it should be faster. Now, now
[6:31] that you're set up, you also want to
[6:33] make sure your web interface is working.
[6:35] So when you first installed it, you
[6:37] should have an address like 127.0.0.1
[6:42] 18789. So let's go to that. We've opened
[6:45] our browser and then we're going to
[6:47] paste in the address. Now it's going to
[6:49] say gateway token missing. When you
[6:51] first install it, it should display a
[6:54] URL with a token equals something. And
[6:58] in my case I need to come to overview go
[7:01] to open gateway token and my gateway
[7:05] token is a llama. I press connect click
[7:08] refresh and once I click refresh
[7:11] I can see that on the top right the
[7:13] health is okay and I'm connected is all
[7:16] green. You can click on overview and see
[7:18] stat is okay. Then if I come to chat,
[7:21] you're going to see that the messages
[7:23] that I've been sending earlier are
[7:25] working. And I can also chat here. So
[7:28] let's give it a try. And it's responded
[7:31] to my high. It's working.
[7:33] Congratulations. You have local llm
[7:36] working with your open claw. I know that
[7:38] a lot of people already installed open
[7:40] claw. And if you're not using lama to
[7:42] install open claw, how do you add lama
[7:45] to your existing open claw? Well, if I
[7:48] come here in my web dashboard, I come to
[7:51] config and then I click on raw, you're
[7:55] going to see the configuration file. And
[7:57] in the configuration file, you can
[7:59] change your configuration file to
[8:02] models, providers, or llama. And then it
[8:05] sets it to quen 3.5 9b. But I have to be
[8:09] honest with you, I hate changing the
[8:11] config file. It's really hard. You keep
[8:14] making mistakes and it doesn't work. So,
[8:16] what's the best way to do it? I'm going
[8:17] to show you two ways to do it. Number
[8:19] one, we're going to use any vibe coding
[8:22] tool you have, OpenAI codeex or Claude
[8:26] Code or Gemini, whatever you have, you
[8:28] can use that to add to your model list.
[8:31] And then number two is to directly tell
[8:33] Open Claw to update your config file to
[8:36] include Lama in your model selection.
[8:40] So, let's go back to our terminal
[8:42] interface. And if I select slash
[8:46] open model picker, it's going to allow
[8:48] me to search. I'm going to type in a
[8:50] lama. Right now, it's only got a llama
[8:52] quen 3.5 9b. Now, let's say it's not
[8:56] even there. How do I add new models to
[8:58] that? So, I'm going to exit this. And
[9:01] you can use whatever you like. I'm going
[9:03] to use claude. And I'm going to bring up
[9:06] our llama. You can see I have GMA 3 4B
[9:10] which I downloaded a long time ago and
[9:12] it's not available in my model list. So
[9:14] let's add that. I'm using open claw menu
[9:17] and new addama
[9:20] 3 to my config get as a model. So what
[9:25] it's going to do is that it's going to
[9:27] search all the files on my computer find
[9:29] the config file and it's going to add to
[9:32] my config file so then they can find my
[9:33] model and it's done. So, it's found the
[9:36] config file and it's added GMA 4B
[9:39] automatically without me manually going
[9:41] in and making a mistake. Let's go open
[9:44] model picker. Okay. And you can see that
[9:47] now it's added GMA 4B. And before this,
[9:51] I realized I needed to restart my
[9:53] gateway for it to recognize. So, what
[9:55] you do is you need to type in open claw
[9:58] gateway restart and it will refresh and
[10:01] reload config. and then you'll be able
[10:03] to find the new model you just added. So
[10:06] that's the easiest way to add your lama
[10:09] models to an existing open claw when you
[10:12] haven't installed it using. And the
[10:15] other way is if you're already connected
[10:16] to a cloud platform and you can chat
[10:19] with open claw. What you can do is you
[10:21] can say can you add gamma 3 to my config
[10:24] file so I can choose as a model. Now
[10:27] that will work too but sometimes I
[10:29] realize it does funky things. The best
[10:31] way is to use Claude Codeex or Gemini,
[10:35] whatever AI coding agent you have to do
[10:37] it, but this also works too. So you can
[10:39] try that. And then the last way is to
[10:42] just go in into your config file and
[10:44] modify the configuration file manually.
[10:47] Now that we have open claw installed,
[10:48] this is the hardest part, choosing the
[10:50] right LLM for Lama. And the trick is you
[10:53] have to download another software
[10:55] because in Olama it doesn't give you
[10:57] much information. So download LM Studio,
[11:01] try out different model and then once
[11:03] you find the best model, run it in O
[11:05] Lama because Lama works better with Open
[11:08] Claw. It's got a smaller footprint and
[11:10] it just runs better. So go to
[11:12] lmstudio.ai and then download the app.
[11:15] And once you have it, click on the
[11:17] search icon, click on the discover icon,
[11:20] and you'll see all this great
[11:22] information. Compared to Lama, which
[11:25] only gives you the model name, LM Studio
[11:28] gives you a lot more. Most importantly,
[11:29] it gives you a best match. So, it
[11:31] recognizes your computer's
[11:32] specifications and then recommends the
[11:35] best one for you. What you have to pay
[11:37] attention to is two things. Number one,
[11:39] how big is the model? So, the larger the
[11:42] model size, the more powerful it is, but
[11:45] also the longer it takes to run. So,
[11:47] it's recommending a lot of smaller
[11:48] models for me, like 9B, 1.2 billion. But
[11:52] let's look at the most downloads. You
[11:54] have ones are 20 billion, 30 billion,
[11:57] and that's just too much for my computer
[11:59] to run. The second thing you should look
[12:01] for is you probably want to look for a
[12:04] model that is compatible with tool use
[12:06] because it means it's designed for agent
[12:09] use like Open Claw. So, pick models that
[12:12] have this symbol on it and then just
[12:14] click download. And after you've
[12:16] selected it, you need to give it a test
[12:17] run to test it speed and its
[12:19] performance. So, let's give it a try.
[12:21] I'm going to load a model and right now
[12:24] I have Quen 4B
[12:27] and I'm going to type test. Okay, so it
[12:30] came back and you can see that this took
[12:32] about 10 seconds and for me this is a
[12:36] little bit slow for me. I want something
[12:38] faster. So I'll test a different model.
[12:41] So keep testing to find the one where
[12:44] the speed is good for you and also
[12:47] you're happy with the results it's
[12:49] giving you. All right, I've been testing
[12:50] a lot of models and I made another video
[12:52] on setting up your own local AI, but
[12:55] here's what I found. It's basically a
[12:57] balance between speed and performance.
[13:00] So, with Open Claw, a lot of people say
[13:02] Kimmy K 2.5 is really good. It is really
[13:05] good. The results that it comes back
[13:07] with is really good, but my computer is
[13:10] not fast enough to run it. So, I get
[13:12] super slow response speeds. Then I tried
[13:14] using LFM2, which is super fast and
[13:18] lightweight. is one of the smallest and
[13:20] fastest models out there. But the
[13:22] results that it gives me are not very
[13:25] smart and I've switched to the current
[13:28] winner is Quen 3.59B. In terms of speed
[13:31] and performance is the best one so far
[13:33] on the market. But the key is you have
[13:36] to keep constantly updating. I was using
[13:38] Quen 3 6 months ago and now the model
[13:42] has improved so much. So, the key is to
[13:46] check for new models every month or two,
[13:49] play around with it in LM Studio, and
[13:51] then strike a balance between speed and
[13:54] quality. And I'm excited because the
[13:56] open- source models you can download are
[13:58] getting much better very quickly. Okay,
[14:00] now we're going to go into a more
[14:01] advanced setup. And this is my setup. I
[14:04] have a Jetson Nano, kind of like a
[14:06] Raspberry Pi, running my Open Claw. But
[14:09] although it is a cheap device, it's not
[14:12] powerful enough to run AI models on it.
[14:14] So I can only run open claw. So I use my
[14:17] old gaming laptop and on that I install
[14:20] llama and I run my AI model on that. Why
[14:23] don't I just run open claw on my gaming
[14:25] laptop as well? Well, because my gaming
[14:28] laptop is not designed to be run 24/7.
[14:31] And I do want it to run 24/7 because I
[14:34] want to message it at any time and I
[14:36] want it to run tasks overnight. is not
[14:39] designed for that. The Jetson Nano is.
[14:41] So, I have both computers running at
[14:42] home and this only turns on when I need
[14:45] to run AI models on it. And why the
[14:47] Jetson Nano? Well, I wouldn't recommend
[14:49] it. It's just something that I have
[14:51] lying around. The best option is
[14:53] actually a Raspberry Pi. So, you can get
[14:55] an $80, $150 option out there where you
[15:00] can buy a Raspberry Pi, buy a nice
[15:02] little case, and you can leave it on
[15:04] running with low electricity costs 24/7,
[15:07] no problem. or you can get a Mac Mini.
[15:09] More expensive, but you can even run
[15:11] your LL models allin-one. So, it depends
[15:14] on your budget. Enough talk. I'm going
[15:15] to show you my setup with a Jetson Nano
[15:18] running Open Claw and having my old
[15:20] gaming laptop as a Lama LLM server
[15:24] connecting to each other. Here's my
[15:26] Jetson Nano. I bought a little case and
[15:30] it even shows the temperature, the CPU
[15:32] usage, the RAM.
[15:35] That's my Jessen Nano running 24/7.
[15:37] Here's my old gaming laptop. It's
[15:40] running. It's on. And the screen isn't
[15:43] even on to conserve energy. And I'm
[15:46] plugged it in so that it can wake on
[15:49] LAN. I'm on my Windows machine right
[15:51] now, which will act as a server to my
[15:54] JSON Nano. And I've gone to the website.
[15:57] I've copied the command, which is this
[16:00] command. And now I'm installing it on my
[16:03] Windows machine. So, I typed in
[16:05] terminal, open terminal, and now it's
[16:08] installing OAMA. And so, I've installed
[16:10] it. And the next thing you need to do is
[16:12] to download a model. So, all you need to
[16:15] do is put in this command, Alama run,
[16:18] and then copy the model name and put it
[16:21] in. So, you can go to models. We're
[16:23] running Quen 3.59B. Copy that and then
[16:26] paste it in and then hit enter. And it
[16:30] will download the model and also run it
[16:32] at the same time. Okay. And it's done.
[16:35] So, let's give it test. And it's work.
[16:37] Okay. And it's working. So, I'm going to
[16:40] just exit this for a little sec. I'm
[16:43] going to exit this. Now, this is only
[16:46] running on my Windows, but in order for
[16:49] it to be a server, I need to allow other
[16:52] computers to access my Windows. So,
[16:55] first of all, I need to type in IP
[16:56] config. And you'll see that my IP
[16:59] address is 1 192.168.68.62.
[17:05] So I need to remember this address.
[17:06] Going to copy it. Then you have to run a
[17:08] command calledama serve. Now you'll see
[17:12] that right now it's serving on what's
[17:15] called localhost 11434,
[17:18] port 11434. And even though this is on,
[17:21] it doesn't mean that my other devices on
[17:24] the network can access it. So, I'll need
[17:27] to set the IP to the IP we just
[17:29] discovered earlier. So, I'm going to
[17:32] come out of here and I'm need to type in
[17:34] this command dollar sign envah host
[17:37] equals and then we're going to put in
[17:38] the address we did earlier. Okay, I put
[17:40] in the address and then now I type in
[17:44] serve and the host is now at this
[17:46] address. So from my other devices, my
[17:49] Justin Ano, I can now call this address
[17:52] and it should be able to communicate
[17:54] with it and use this as my server for my
[17:58] open call. But every time you turn off
[18:00] the computer and you restart it, your IP
[18:03] address may change. So I'm going to show
[18:06] you how I set a static IP on my router.
[18:09] And there's more. I'm gonna put this on
[18:12] wake on land. So then my computer can
[18:15] turn off when it's not being used and
[18:17] then only when it's being called by my
[18:19] open call, it'll wake up, turn on the
[18:21] power and then run the llama. Every time
[18:23] my gaming laptop shuts down and then
[18:26] powers on again, it will get a new IP
[18:28] address. And I don't want that because
[18:30] if my Justin Nano is calling it, we want
[18:32] a fixed address so it knows exactly
[18:34] where it is. So depending on your router
[18:36] settings, this is my router. I use TP
[18:41] link. What I do is I can come to more. I
[18:44] come to advanced.
[18:47] I go to address reservation.
[18:51] And basically I've reserved my device to
[18:55] always have a static IP on 192.168
[19:00] 6865.
[19:01] And so it's fixed and that's reserved
[19:04] for my Alama server running on my gaming
[19:08] laptop. So depending on your router
[19:10] settings, you need to set a static IP.
[19:12] And that's how you set it. Now I'm going
[19:13] to show you how to set some settings. So
[19:15] then you can have the computer turn off
[19:18] and then wake on land. So it only turns
[19:21] on when you actually need to use it. I'm
[19:24] going to go and restart my
[19:27] Windows.
[19:29] And then while it's restarting, just
[19:31] keep tapping the delete button to bring
[19:34] up the BIOS. Okay, now the BIOS is
[19:37] brought up. You want to go to advanced.
[19:40] You want to enable USB power and sleep
[19:43] in hibernation. You want to disable
[19:47] fast boot and then you want to enable
[19:49] wake on land and then you save and
[19:52] reset. Now depending on your machine,
[19:55] you might have a different BIOS. So
[19:57] check with AI what settings you need,
[20:00] but generally that's what you need to
[20:02] enable wake on land. One more thing,
[20:04] just make sure you use an Ethernet cable
[20:08] so you have a cable plugged in because
[20:12] it doesn't wake up if it's just on
[20:14] Wi-Fi. Now that I have Alama as a
[20:17] server, there's a quick little tip. By
[20:20] default, the context length might be too
[20:23] small. And so, you need to go and type
[20:26] in a command that says show and then
[20:28] quen 3.59B and it will show you your
[20:32] context length. So right now it's at
[20:34] 26,000 2,144.
[20:37] So I've increased it, but you might just
[20:40] see 4,000 here. And if that's the case,
[20:42] you need to increase it to a
[20:44] recommendation of at least 16,000. The
[20:46] way to set the context length is to
[20:49] enter run your model name dash context
[20:54] and then enter your number. Play around
[20:55] with the number. Obviously, the larger
[20:58] the context length, the better. But
[21:00] also, if you have a really long request,
[21:03] it might crash your computer. So, play
[21:05] around with this number. I've set mine
[21:07] to 26,000. At the minimum you should set
[21:10] it to is around 16,000. Another thing
[21:12] that I did was that once I've got the
[21:14] llama running, I don't want my gaming
[21:16] computer to be on all the time because
[21:19] it's not designed that way and it'll
[21:20] overheat really quickly. So, what I've
[21:23] done is I've gone into my system power
[21:26] and battery. I'm going to turn off my
[21:28] screen after 5 minutes of nothing going
[21:30] on. And then it will hibernate after 10
[21:32] minutes. And if it goes to sleep, that's
[21:34] fine. We're going to wake on land. So
[21:37] then when my other devices call this
[21:40] device, it will wake up. So how do you
[21:42] do that? Well, you have to come to your
[21:44] device manager and you have to look for
[21:46] your internet connection devices. So I
[21:50] have my Wi-Fi and I also have a LAN
[21:52] connection where you plug in a cable. So
[21:55] this is my Wi-Fi.
[21:57] So if I want to wake my Wi-Fi, I can
[21:59] come to power management and allow this
[22:02] device to wake the computer and press
[22:04] okay. Enable that. And the other one is
[22:08] my LAN cable and I do the same. And you
[22:10] can find this uh inside your network
[22:13] adapter. And once you do that, the last
[22:15] remaining thing you can do is to restart
[22:18] your computer and in your BIOS set wake
[22:22] on land. And one final thing is that you
[22:25] don't have to set a static IP. There's
[22:27] an even more advanced way to set this up
[22:29] and that's by using tail scale. By using
[22:31] tails scale, not only is it more secure,
[22:33] but you can also access this server from
[22:36] any device even when you are outside of
[22:39] your house. In my current setup,
[22:42] everything has to happen in the same
[22:43] network and you have to be on the same
[22:45] Wi-Fi to work. But let's say I'm out and
[22:48] about and I want to access my Olama
[22:50] server. I can do that using tail scale
[22:53] and tunnel in and it's even more secure,
[22:56] but that's for another video. So stay
[22:58] tuned for my other video. I'm back on my
[23:00] Mac. I have open claw installed. And
[23:02] then we have a lama set up on my old
[23:05] gaming laptop on a separate machine. So
[23:07] the first thing we need to do is to
[23:09] check if it can connect to it. You can
[23:11] do that by using a curl command in your
[23:13] terminal. So open your terminal, type in
[23:16] this command, curl http your IP address
[23:19] of your gaming machine/ ai/tags and it's
[23:23] returned that it's got a model 3.59b. So
[23:26] it's working. Okay, now that it's
[23:28] connected to it, let's see if it
[23:29] actually runs. So the next command is
[23:32] curl API generate d choose a model. Say
[23:37] my prompt is hello and I want to choose
[23:38] stream. So I'll include this prompt
[23:41] inside the description and you can see
[23:43] that it's coming back with some response
[23:45] and it's thinking. It's thinking. Let's
[23:47] check back in when it gives me a result.
[23:49] Okay. And it's done and it's giving me a
[23:51] response. So it's working. So now we're
[23:53] ready to go to open claw. So let's go to
[23:56] open claw tui. So now that my MacBook
[23:59] can connect to the server. There's three
[24:01] ways to change the configuration
[24:03] settings. So then open claw can use the
[24:06] model remotely. Number one, you can go
[24:09] into config file and paste this into it
[24:11] and set your provider as a llama, your
[24:14] IP address and the model. You can do
[24:16] that manually, but every time I've done
[24:18] that, it has not been successful. But if
[24:20] you really want to do it, all you need
[24:21] to do is go to finder, go to your home
[24:24] directory, and then go command, shift,
[24:27] and period. And you're going to see all
[24:29] the hidden files. And in there, I'm
[24:31] going to find open claw. So let's find
[24:33] open claw openclaw.json. Let's open it.
[24:36] And here I can input my I can copy this
[24:40] the model provider and paste it into
[24:43] here. But so far every time I've done
[24:45] it, it has not been successful. So I'm
[24:48] going to do in my opinion the best
[24:50] option which is to use vibe coding to do
[24:52] it. So in this case I'm going to use
[24:53] cloud code. You can use openai Gemini.
[24:56] I'm going to choose a folder and I'm
[24:59] going to go back into the same folder.
[25:01] I'm going to go back into open claw.
[25:03] Okay. And then I'm just going to say
[25:05] modify the open claw config file. So
[25:08] then it points to my remote or llama
[25:11] server. And then you plug in the
[25:14] details, your IP address and the model.
[25:17] And then it should automatically make
[25:19] those changes. Okay, it says it's done.
[25:22] And so I need to come back. And then the
[25:24] thing I need to do is open call gateway
[25:27] restart. So, I need to restart the
[25:28] gateway to activate the change and then
[25:32] let's test it out. And there we have it.
[25:34] It's returned a response. Hi there, I'm
[25:36] still awake. Let's start with the
[25:38] basics. So, it's working right now. And
[25:40] the third option is a chicken and egg
[25:42] problem. You can ask open claw to update
[25:45] the model similar to how I did it in
[25:47] cloud code. Just tell it that you have a
[25:50] new remote server. Give it the IP and
[25:52] the model and it will update it. But the
[25:55] chicken egg problem is that you need an
[25:56] AI LLM provider connected to do this. So
[25:59] either connect to your cloud service to
[26:01] do this or you can run a really small
[26:04] model on your local computer just so
[26:07] then you can get the configuration done
[26:08] so then openclaw can set its own config
[26:11] file. So those are the three options you
[26:12] can use to connect and change your model
[26:16] to point to a remote server. So that's
[26:18] how you run open claw locally. We
[26:20] started with the beginner setup where
[26:22] you're running openclaw on just one
[26:24] machine and then I showed you my current
[26:25] setup where a small computer runs open
[26:28] claw 24/7 and another machine runs AI
[26:32] model. But the bigger question is should
[26:34] you run open claw locally at all? If
[26:37] you're an absolute beginner honestly the
[26:39] best option is to run open claw on a
[26:42] virtual private server. So, you rent a
[26:44] server somewhere and then call a cloud
[26:46] model like OpenAI, Gemini, or Claude
[26:48] using their API. There are ton of
[26:50] one-click services out there for Open
[26:52] Claw and you'll spend much less time
[26:54] configuring things. But if privacy
[26:56] matters to you or you're worried about
[26:58] the API cost, then running Open Claw
[27:00] could be a very powerful option. Your
[27:02] data stays on your machine and your
[27:04] system keeps running even if the cloud
[27:06] goes down. And the number one
[27:08] requirement is that you need a computer
[27:10] that's fast enough to run a local LLM.
[27:13] In my case, I'm just using an old gaming
[27:15] laptop, and that works surprisingly
[27:17] well. The trade-off is that local setups
[27:20] require more work, networking, hardware,
[27:23] testing different local LLM models until
[27:25] you find the right balance between speed
[27:28] and quality. But once everything is
[27:30] running, you basically have your own AI
[27:32] infrastructure at home. And that's
[27:34] pretty sweet. Local AI gives you the
[27:36] control, but it also makes you the
[27:38] system administrator. If you've liked
[27:40] this video, please like and subscribe to
[27:42] my channel. If you want to learn more
[27:43] about AI, you can join my free AI
[27:45] community in the description. I look
[27:46] forward to seeing you