Mac Mini For AI Agents: How Much Memory You Need, And How To Use What You Have.

I'll just say it I'll just say it straight. You are no longer the straight. You are no longer the straight. You are no longer the only user of only user of only user of your computer. your computer. your computer. If you run If you run If you run agents all day, agents all day, agents all day, they probably they probably they probably use it use it use it more than you do. Now I more than you do. Now I more than you do. Now I know what you're know what you're know what you're thinking. My Mac is getting old thinking. My Mac is getting old . I need a more . I need a more . I need a more powerful one. Maybe powerful one. Maybe powerful one. Maybe you're watching this video you're watching this video you're watching this video because you're looking at a Mac because you're looking at a Mac because you're looking at a Mac mini or a Mac Studio, or maybe you're mini or a Mac Studio, or maybe you're mini or a Mac Studio, or maybe you're comparing some comparing some comparing some computer with Nvidia. That's the computer with Nvidia. That's the computer with Nvidia. That's the problem. A problem. A problem. A faster chip and more faster chip and more faster chip and more memory won't stop the memory won't stop the memory won't stop the increasing increasing increasing load from the load from the load from the agents you agents you agents you run. Your run. Your run. Your agents open agents open agents open browsers. They browsers. They browsers. They launch services. launch services. launch services. They write files. They write files. They write files. And after they finish their And after they finish their And after they finish their job, they job, they job, they leave a lot of leave a lot of leave a lot of trash on your trash on your trash on your car. Essentially, you car. Essentially, you car. Essentially, you have a messy neighbor, and no have a messy neighbor, and no one has set any rules for living there. one has set any rules for living there. So we have to So we have to So we have to install them together install them together install them together right here. Let's go. And don't right here. Let's go. And don't right here. Let's go. And don't worry, I've already worry, I've already worry, I've already written these written these written these rules of rules of rules of cohabitation for you. This cohabitation for you. This cohabitation for you. This video is accompanied by a video is accompanied by a video is accompanied by a guide on Substack with guide on Substack with guide on Substack with three requests three requests three requests you can give you can give you can give directly to your directly to your directly to your agent. The first is to agent. The first is to agent. The first is to find what's eating up find what's eating up find what's eating up your memory, the second your memory, the second is to clean only is to clean only is to clean only what you approve, and what you approve, and what you approve, and the third is to the third is to the third is to set up set up set up regular checks regular checks regular checks so that so that so that memory usage and anything that memory usage and anything that memory usage and anything that slows down your slows down your slows down your computer doesn't computer doesn't computer doesn't build up again build up again . The link is in the description . The link is in the description . The link is in the description below, and if you're below, and if you're below, and if you're connected to connected to connected to Nate's MCP, your agent can Nate's MCP, your agent can Nate's MCP, your agent can just grab these just grab these just grab these requests directly requests directly requests directly from the library and from the library and from the library and get started right now. You get started right now. You get started right now. You don't even need to don't even need to don't even need to go to Substack. You go to Substack. You go to Substack. You can just can just can just use MCP. So, use MCP. So, use MCP. So, by the end of this video, by the end of this video, by the end of this video, you'll know which you'll know which you'll know which computer is right computer is right computer is right for for for your agent workloads, how to your agent workloads, how to your agent workloads, how to get more out get more out get more out of the machine you have of the machine you have of the machine you have on your on your on your desk right now, and how to ensure desk right now, and how to ensure desk right now, and how to ensure your your your agent experience agent experience agent experience remains robust remains robust remains robust even as even as even as new tasks are added. I new tasks are added. I new tasks are added. I will use will use will use four principles to four principles to four principles to connect these solutions connect these solutions . Let's start with a "medical checkup . Let's start with a "medical checkup " of your Mac. If you " of your Mac. If you " of your Mac. If you open open System Monitor on your Mac

System Monitor on your Mac System Monitor on your Mac and click on the and click on the and click on the Memory tab, you Memory tab, you Memory tab, you can literally can literally can literally see a graph of the see a graph of the see a graph of the load while load while load while your agents your agents your agents are doing important are doing important are doing important work. And it's work. And it's work. And it's colored, right? colored, right? colored, right? Green means Green means Green means everything is fine. Yellow and everything is fine. Yellow and everything is fine. Yellow and red mean the red mean the red mean the car is barely car is barely car is barely coping. To be coping. To be coping. To be honest, I honest, I honest, I often have them yellow or often have them yellow or often have them yellow or red. See red. See red. See which apps which apps which apps are using are using are using memory. A high memory. A high RAM usage rate RAM usage rate doesn't necessarily mean you doesn't necessarily mean you need a new need a new computer, but it's computer, but it's computer, but it's definitely a signal that it's definitely a signal that it's definitely a signal that it's worth checking your worth checking your worth checking your system. Then, if system. Then, if system. Then, if you continue your you continue your you continue your review, check review, check review, check the drive. the drive. the drive. Look at Look at Look at system system system settings, settings, settings, general, and storage. general, and storage. general, and storage. And RAM And RAM is the workspace is the workspace is the workspace for what you're doing for what you're doing . So, the repository . So, the repository . So, the repository will contain your files will contain your files will contain your files and your and your and your software. macOS software. macOS software. macOS can use can use can use storage for a storage for a storage for a paging file, which paging file, which paging file, which temporarily stores temporarily stores temporarily stores the contents of memory so that the contents of memory so that the contents of memory so that they don't interfere with each they don't interfere with each they don't interfere with each other. But essentially, other. But essentially, other. But essentially, you need to you need to you need to think about how you think about how you think about how you allocate that allocate that allocate that memory so that you're not memory so that you're not memory so that you're not trying to build trying to build trying to build on top of a bunch of agents in on top of a bunch of agents in on top of a bunch of agents in different environments. And different environments. And different environments. And that's one of the things I that's one of the things I that's one of the things I notice, for example, notice, for example, notice, for example, when you have when you have when you have Claude open in Chrome and another Claude open in Chrome and another Claude open in Chrome and another browser is taking up browser is taking up browser is taking up a lot of a lot of a lot of RAM. I RAM. I RAM. I use Zen. When use Zen. When use Zen. When you use you use you use a lot of memory in Codex and a lot of memory in Codex and a lot of memory in Codex and even more in Claude, and even more in Claude, and even more in Claude, and then add another then add another then add another AI tool, AI tool, AI tool, like for like for like for transcription, it all transcription, it all transcription, it all adds up. This creates a adds up. This creates a adds up. This creates a load on load on load on memory. This is something that memory. This is something that memory. This is something that laboratories have already laboratories have already laboratories have already paid attention to. Thibault paid attention to. Thibault paid attention to. Thibault from OpenAI recently from OpenAI recently from OpenAI recently released an update released an update released an update to Codex that reduces to Codex that reduces to Codex that reduces memory consumption.

memory consumption. memory consumption. This was definitely a This was definitely a This was definitely a problem. But these problem. But these problem. But these point fixes point fixes point fixes from the labs are from the labs are from the labs are n't enough, n't enough, n't enough, especially if, like especially if, like especially if, like most of us, you most of us, you most of us, you run multiple run multiple run multiple workflows. I workflows. I workflows. I know many in the know many in the know many in the community here community here community here run local run local run local models like Quen, as well models like Quen, as well models like Quen, as well as models as models as models like Claude or Codex. We like Claude or Codex. We like Claude or Codex. We need a system that need a system that need a system that works with all of them. And that's exactly what it's about works with all of them. And that's exactly what it's about works with all of them. And that's exactly what it's about . . . This brings us to the This brings us to the This brings us to the first principle. You first principle. You first principle. You need to know where need to know where need to know where the work is happening. the work is happening. the work is happening. Let's say you Let's say you Let's say you ask OpenAI Codex or ask OpenAI Codex or ask OpenAI Codex or Anthropic Claude to create a Anthropic Claude to create a Anthropic Claude to create a small website for you small website for you small website for you . A very typical . A very typical . A very typical request, isn't it? request, isn't it? request, isn't it? The model may The model may The model may respond from the data respond from the data center, but that doesn't center, but that doesn't center, but that doesn't mean there is no mean there is no mean there is no load on your load on your load on your system. Your system. Your system. Your computer is still computer is still computer is still running the running the running the development server, the development server, the development server, the site validation browser, and the site validation browser, and the site validation browser, and the build tools. It's build tools. It's build tools. It's all local, and you all local, and you all local, and you feel it, right feel it, right ? If you give another ? If you give another ? If you give another agent a different project agent a different project agent a different project at the same time, it at the same time, it at the same time, it will run a different will run a different will run a different set of tools, and it will all hit set of tools, and it will all hit your local your local computer. I experienced computer. I experienced computer. I experienced this when I this when I this when I was using Claude was using Claude was using Claude to create videos in to create videos in to create videos in Blender while Blender while Blender while having Codex having Codex having Codex do other do other do other work. It just " work. It just " eats" your memory.

eats" your memory. eats" your memory. Therefore, if you Therefore, if you Therefore, if you receive an receive an out of memory warning, out of memory warning, consider local consider local consider local tools during tools during tools during diagnostics and diagnostics and diagnostics and expect expect expect them to be the them to be the them to be the primary path to primary path to primary path to resolution. This doesn't resolution. This doesn't resolution. This doesn't mean you mean you mean you should stop should stop should stop working with agents. We're working with agents. We're working with agents. We're not saying this, but not saying this, but not saying this, but you have to think about you have to think about you have to think about how you do it and how you do it and how you do it and delve into the details. delve into the details. delve into the details. You need to You need to You need to make sure you don't make sure you don't make sure you don't just buy just buy just buy a graphics card a graphics card a graphics card hoping for a one- hoping for a one- hoping for a one- time solution to time solution to time solution to this problem. As I this problem. As I this problem. As I keep saying, you can keep saying, you can keep saying, you can buy more buy more buy more power, but you power, but you power, but you can't can't can't buy away the buy away the buy away the problem itself, because problem itself, because problem itself, because agents will become agents will become agents will become increasingly capable. increasingly capable. increasingly capable. So, the main So, the main So, the main conclusion: treat the conclusion: treat the conclusion: treat the memory problem as a memory problem as a memory problem as a local one that should be local one that should be local one that should be solved with agents solved with agents solved with agents before adding before adding before adding new ones. If you new ones. If you new ones. If you are running a model are running a model are running a model locally, you should locally, you should locally, you should consider the consider the consider the computational requirements computational requirements computational requirements for that model. And for that model. And for that model. And if you have a mixed if you have a mixed if you have a mixed system, which I system, which I system, which I think is common for think is common for think is common for many, you should many, you should many, you should understand by looking understand by looking understand by looking at the settings at the settings at the settings I mentioned what exactly is I mentioned what exactly is I mentioned what exactly is loading memory loading memory loading memory from a particular from a particular from a particular application. When Qwen application. When Qwen application. When Qwen loads your loads your loads your local memory, and local memory, and local memory, and when Codex and Claude do, these are when Codex and Claude do, these are when Codex and Claude do, these are different things. You different things. You different things. You should be able to should be able to should be able to see it. Let's see it. Let's see it. Let's go back to our go back to our go back to our website story. website story. website story. Imagine that the agent Imagine that the agent Imagine that the agent has completed a site, you has completed a site, you has completed a site, you have moved on to another one, have moved on to another one, have moved on to another one, but the development server but the development server but the development server and browser are still and browser are still and browser are still open in the open in the open in the background. People don't background. People don't background. People don't realize that their realize that their realize that their attention attention attention has already switched, but these has already switched, but these has already switched, but these tools are still tools are still tools are still consuming resources in a consuming resources in a consuming resources in a passive mode.

passive mode. passive mode. They still hold They still hold They still hold your memory your memory your memory hostage, like hostage, like hostage, like blocked funds on a blocked funds on a blocked funds on a card. The project has card. The project has card. The project has dependencies, dependencies, dependencies, build files that occupy build files that occupy build files that occupy your repository. You your repository. You your repository. You need to actually need to actually need to actually close everything if you close everything if you close everything if you want these files to want these files to want these files to disappear or to disappear or to disappear or to free up working free up working free up working memory. And you might memory. And you might memory. And you might think Nate think Nate think Nate is talking mostly is talking mostly is talking mostly about agents, but about agents, but about agents, but honestly, part of the honestly, part of the honestly, part of the problem comes problem comes problem comes from bugs in the from bugs in the from bugs in the agent agent software. In software. In one public one public one public Codex report, a user Codex report, a user Codex report, a user described a described a described a crash report folder that had crash report folder that had crash report folder that had grown to over grown to over grown to over 50GB just describing these 50GB just describing these 50GB just describing these errors. Another errors. Another errors. Another user described user described user described over 100GB of Git objects over 100GB of Git objects associated with associated with session checkpoints. These are specific session checkpoints. These are specific cases. I'm not saying this is cases. I'm not saying this is cases. I'm not saying this is your problem per se your problem per se your problem per se . I say . I say . I say pay attention to the pay attention to the pay attention to the details on your details on your details on your computer. Pay computer. Pay computer. Pay attention to your attention to your attention to your specific specific specific configuration. That's configuration. That's configuration. That's why I emphasize the why I emphasize the why I emphasize the audit, and only then audit, and only then audit, and only then can you can you can you develop a strategy with the develop a strategy with the develop a strategy with the agent to agent to agent to address it. This is address it. This is address it. This is absolutely useful absolutely useful absolutely useful work for the agent, because work for the agent, because work for the agent, because the short version of the the short version of the the short version of the diagnostic diagnostic diagnostic query that you give to query that you give to query that you give to the agent looks like this the agent looks like this the agent looks like this , right? You just , right? You just , right? You just tell him: check tell him: check tell him: check my computer.

my computer. my computer. Please don't Please don't Please don't change anything. Please change anything. Please change anything. Please identify the largest identify the largest identify the largest memory consumers, those memory consumers, those memory consumers, those that "eat" it, and the that "eat" it, and the that "eat" it, and the corresponding corresponding corresponding directories. Explain directories. Explain directories. Explain what what what each of them refers to and each of them refers to and each of them refers to and suggest specific suggest specific suggest specific fixes that fixes that fixes that will help me in will help me in will help me in my work. my work. my work. I now have the full text of the I now have the full text of the I now have the full text of the Substack request and a Substack request and a Substack request and a guide with guide with guide with more detailed more detailed more detailed information, because what information, because what information, because what you really you really you really need is a comprehensive need is a comprehensive system “hygiene” program. But in fact, it system “hygiene” program. But in fact, it 's worth starting with 's worth starting with 's worth starting with such a report, because just because a such a report, because just because a such a report, because just because a folder is folder is folder is being used being used being used doesn't mean it's being doesn't mean it's being doesn't mean it's being used for used for used for good. We need to good. We need to good. We need to delve into the details, delve into the details, delve into the details, right? For example, right? For example, right? For example, let's say you're let's say you're let's say you're doing a test doing a test doing a test and discover that you have a and discover that you have a bunch of bunch of development servers running at the same time. Hey, it's me. development servers running at the same time. Hey, it's me. development servers running at the same time. Hey, it's me. I've done this for I've done this for I've done this for several of my several of my several of my projects. Before projects. Before projects. Before stopping one of them, stopping one of them, stopping one of them, find out what it is for find out what it is for find out what it is for . If the project . If the project . If the project is complete, if is complete, if is complete, if the server is no longer the server is no longer the server is no longer needed, was it needed, was it needed, was it just an experiment? just an experiment? just an experiment? Then close it and Then close it and Then close it and see if the see if the memory load decreases. See memory load decreases. See if if if the load on the the load on the the load on the storage decreases. Determine what storage decreases. Determine what storage decreases. Determine what can be restored, can be restored, can be restored, then come up with a then come up with a then come up with a way to clean it up. If way to clean it up. If way to clean it up. If all this sounds really all this sounds really all this sounds really boring, I have news for you boring, I have news for you boring, I have news for you . We've been living with . We've been living with computing computing resource constraints since the resource constraints since the resource constraints since the 1980s, and now we're 1980s, and now we're 1980s, and now we're back to it again, back to it again, back to it again, because agents "live" with us on these because agents "live" with us on these because agents "live" with us on these computers computers computers , and we just , and we just , and we just need to learn to be need to learn to be need to learn to be good neighbors.

good neighbors. good neighbors. Another example Another example Another example worth paying attention to is worth paying attention to is problems with problems with problems with SSD memory. SSD memory. SSD memory. Solid-state Solid-state Solid-state drives are very drives are very drives are very fast to write and fast to write and fast to write and rewrite. Earlier rewrite. Earlier rewrite. Earlier this year, Codex this year, Codex this year, Codex had an issue with had an issue with had an issue with excessive logging excessive logging . One . One . One user user user recorded about 37 recorded about 37 recorded about 37 terabytes of writes terabytes of writes terabytes of writes on an SSD in 21 days. That's a on an SSD in 21 days. That's a on an SSD in 21 days. That's a lot. lot. Logging fixes were Logging fixes were released by Codex in June, released by Codex in June, released by Codex in June, and a later release and a later release and a later release added the ability to added the ability to added the ability to free up free up free up unused unused log database space. This is progress, but log database space. This is progress, but at the same time it reminds at the same time it reminds at the same time it reminds me and should remind me and should remind me and should remind you that this is you that this is you that this is software that is software that is software that is evolving, and we evolving, and we evolving, and we need to keep a close eye need to keep a close eye need to keep a close eye on how on how on how it uses the it uses the it uses the physical components of physical components of physical components of our machine. And, our machine. And, our machine. And, of course, there is also of course, there is also of course, there is also growth that growth that growth that occurs simply because occurs simply because occurs simply because agents become agents become agents become more capable. more capable. more capable. More useful More useful More useful features work as features work as features work as intended. intended. intended. Look at the Look at the Look at the progress we've progress we've progress we've seen in seen in coding agents since January. You coding agents since January. You can create can create can create working trees, have working trees, have working trees, have separate working separate working separate working directories. A directory directories. A directory directories. A directory can have its own can have its own can have its own dependencies and dependencies and dependencies and build caches. And Codex has build caches. And Codex has build caches. And Codex has settings to settings to settings to manage all of these manage all of these manage all of these environments. And before environments. And before environments. And before you choose one of them you choose one of them you choose one of them and decide on and decide on and decide on the settings, it's the settings, it's the settings, it's worth thinking carefully about worth thinking carefully about worth thinking carefully about how and where you want to how and where you want to how and where you want to carry out this carry out this carry out this assembly project. And here I assembly project. And here I assembly project. And here I will make a small will make a small will make a small digression about digression about digression about using cloud using cloud using cloud and on-premises and on-premises and on-premises environments. I'll open environments. I'll open environments. I'll open the cards. I the cards. I the cards. I like the local like the local like the local environment, while it's environment, while it's environment, while it's possible, because it possible, because it possible, because it makes it very easy makes it very easy makes it very easy to switch between to switch between to switch between tools and AI tools and AI tools. I tools. I tools. I can just point can just point can just point to a file directory and to a file directory and to a file directory and say: say: say: use this. So there's use this. So there's no escaping the need to think things through no escaping the need to think things through . Regardless of . Regardless of . Regardless of whether whether whether you use you use you use the cloud or an on-premises the cloud or an on-premises the cloud or an on-premises environment. My environment. My environment. My main rule here is to main rule here is to give every agent the

give every agent the give every agent the "stop" sign. Give "stop" sign. Give "stop" sign. Give him something that will make him something that will make him something that will make him finish the him finish the him finish the job. And then job. And then job. And then put the results where put the results where put the results where I can I can I can find them. Of course? This is something to find them. Of course? This is something to think about, because I think about, because I think we often think we often think we often start things and start things and start things and leave them leave them leave them unfinished. They unfinished. They unfinished. They may be may be may be incomplete in the incomplete in the incomplete in the sense that a Chrome tab sense that a Chrome tab sense that a Chrome tab with your working with your working with your working project is open. project is open. project is open. They may be They may be They may be incomplete in the incomplete in the incomplete in the sense that the chat is sense that the chat is sense that the chat is active and active and active and some some some development server is running. But ultimately development server is running. But ultimately , if you don't have a , if you don't have a , if you don't have a reason to stop, reason to stop, reason to stop, all AI agents all AI agents all AI agents trained through trained through reinforcement learning reinforcement learning will keep will keep will keep working. So they working. So they will just continue to " eat up" the resources of eat up" the resources of eat up" the resources of your local your local your local machine. More machine. More machine. More power gives you power gives you power gives you more space, but it more space, but it more space, but it doesn't solve this doesn't solve this doesn't solve this problem. You'll problem. You'll problem. You'll have to have to have to figure this out on your figure this out on your figure this out on your own. So, that's own. So, that's own. So, that's principle number two. The principle number two. The principle number two. The third principle when third principle when third principle when you're working with all of you're working with all of you're working with all of this is this is this is concurrent concurrent concurrent load management. load management. load management. Let's say you have a Let's say you have a Let's say you have a dozen documents dozen documents dozen documents to process and you to process and you queue 12 tasks. This is quite queue 12 tasks. This is quite realistic. That's what I realistic. That's what I realistic. That's what I do, right? And you do, right? And you do, right? And you run one or run one or run one or two heavy tasks two heavy tasks two heavy tasks at a time. The agent at a time. The agent at a time. The agent is actually able to is actually able to is actually able to analyze this analyze this analyze this workload and workload and workload and plan the entire plan the entire plan the entire task without running task without running task without running every browser, every browser, conversion tool, and conversion tool, and local local local model queries at the same time; model queries at the same time; model queries at the same time; he can do it, and he can do it, and he can do it, and he must he must he must do it. You just do it. You just do it. You just have to ask him about it have to ask him about it have to ask him about it . So, I would . So, I would . So, I would say that in that say that in that say that in that case, you just case, you just case, you just have to say to the agent, have to say to the agent, have to say to the agent, “Can you “Can you “Can you set up to set up to set up to optimize my optimize my optimize my workload? "Start with workload? "Start with workload? "Start with one active one active one active task, maybe task, maybe task, maybe then move to two then move to two , check how , check how , check how the work completes, the work completes, the work completes, check the check the check the memory requirements." Your agent is a memory requirements." Your agent is a great partner in great partner in great partner in this. It is "native" this. It is "native" this. It is "native" to computing. He to computing. He to computing. He can check these can check these can check these things. If adding things. If adding things. If adding tasks slows tasks slows tasks slows everything down, you've realized you've everything down, you've realized you've everything down, you've realized you've stretched your stretched your stretched your current current current computing computing computing resources to the limit and

resources to the limit and resources to the limit and need to slow down a bit need to slow down a bit need to slow down a bit . And that's why . And that's why . And that's why the number of the number of the number of processor cores does not processor cores does not processor cores does not translate directly translate directly translate directly into the into the into the number of agents. There number of agents. There number of agents. There is no simple is no simple is no simple answer here. Two agents answer here. Two agents waiting for waiting for waiting for responses from the cloud responses from the cloud responses from the cloud may may may do almost nothing locally, do almost nothing locally, do almost nothing locally, but two agents but two agents but two agents doing a large doing a large doing a large build or rendering build or rendering build or rendering video will eat up all the video will eat up all the video will eat up all the CPUs you have CPUs you have CPUs you have on hand. So more on hand. So more on hand. So more cores can help cores can help cores can help when the when the when the software is able to software is able to software is able to use them, and use them, and use them, and more memory can more memory can more memory can help when the help when the help when the active workload active workload active workload needs that needs that needs that space, but you can't space, but you can't space, but you can't just just just optimize those optimize those optimize those metrics without thinking metrics without thinking metrics without thinking about it, without thinking about the about it, without thinking about the about it, without thinking about the workload that you're workload that you're workload that you're actually running actually running actually running on that machine. If you on that machine. If you on that machine. If you invest in powerful invest in powerful invest in powerful equipment, you can equipment, you can equipment, you can get more out of get more out of get more out of that equipment using the that equipment using the same process, same process, right? Separate the right? Separate the right? Separate the work stages: let several work stages: let several work stages: let several agents do agents do agents do research and research and research and planning while planning while planning while the large local the large local the large local build is executed one by one build is executed one by one build is executed one by one in a queue. This way in a queue. This way in a queue. This way you can adjust your you can adjust your you can adjust your schedule to your schedule to your schedule to your resources, instead of resources, instead of resources, instead of forcing all forcing all forcing all tasks to compete tasks to compete tasks to compete for them at the same time. And for them at the same time. And for them at the same time. And if you think I'm going to keep if you think I'm going to keep repeating this, you repeating this, you can delegate it to can delegate it to can delegate it to your agents. This is your agents. This is your agents. This is partly why partly why partly why I created my own I created my own I created my own prompts, prompts, prompts, because it's too because it's too because it's too much to handle on my much to handle on my much to handle on my own. Then own. Then own. Then pay attention to the pay attention to the pay attention to the work that doesn't work that doesn't work that doesn't have to have to have to be done, right?

be done, right? be done, right? Have the agent Have the agent Have the agent check the existing check the existing check the existing project before project before project before creating another creating another creating another environment. environment. Reuse the browser where Reuse the browser where appropriate. appropriate. appropriate. Keep your Keep your Keep your work in a work in a work in a predictable predictable predictable folder structure that is folder structure that is folder structure that is understandable to you and understandable to you and understandable to you and your agent. your agent. your agent. Ask him Ask him Ask him to identify processes to identify processes to identify processes that should that should that should continue to continue to continue to run after run after run after completion. And finally completion. And finally , store the , store the , store the knowledge you gain in knowledge you gain in knowledge you gain in project instructions so you project instructions so you project instructions so you can can can use it again. use it again. use it again. If you If you If you 're using 're using 're using local models, local models, software settings matter here, matter here, right? Docker Desktop has right? Docker Desktop has right? Docker Desktop has limitations for its limitations for its limitations for its virtual machine. virtual machine. virtual machine. It has a It has a It has a resource saving function resource saving function resource saving function in standby mode. in standby mode. in standby mode. Model servers Model servers Model servers have have have controls for controls for controls for handling context, handling context, handling context, concurrent requests, concurrent requests, concurrent requests, and the amount of time that and the amount of time that and the amount of time that models models models remain remain remain loaded. All of this loaded. All of this loaded. All of this affects the affects the affects the load on your load on your load on your computer. So in computer. So in computer. So in the guide I'll talk the guide I'll talk the guide I'll talk about Ollama, LM Studio, Docker, and about Ollama, LM Studio, Docker, and about Ollama, LM Studio, Docker, and how to how to how to set it up, because as set it up, because as set it up, because as you might have guessed, I'm you might have guessed, I'm you might have guessed, I'm really keen really keen really keen to get the most out to get the most out to get the most out of your machine. And of your machine. And of your machine. And if you run if you run if you run local models, local models, local models, you know what? You will still you know what? You will still you know what? You will still have to have to have to determine the size of determine the size of determine the size of the request and handle the request and handle the request and handle concurrent work on the concurrent work on the concurrent work on the task. A larger task. A larger task. A larger context will require context will require context will require more memory if more memory if more memory if it is a local model. it is a local model. it is a local model. Multiple queries Multiple queries Multiple queries will be will be will be summed. summed.

summed. Keeping the model Keeping the model Keeping the model constantly constantly constantly loaded can loaded can loaded can reduce reduce reduce the load on the the load on the the load on the next request. This next request. This next request. This reduces latency. But reduces latency. But reduces latency. But it will still it will still it will still consume memory between consume memory between consume memory between tasks. You will tasks. You will tasks. You will have to work have to work have to work with your agent to with your agent to with your agent to choose a compromise choose a compromise choose a compromise depending on how depending on how depending on how you use it. you use it. you use it. This gives us an excuse to This gives us an excuse to This gives us an excuse to talk about purchasing talk about purchasing talk about purchasing equipment. When equipment. When equipment. When the job requires the job requires the job requires more power, more power, more power, you have exhausted you have exhausted you have exhausted the capabilities of your the capabilities of your the capabilities of your machine, conducted an audit machine, conducted an audit machine, conducted an audit and checked the and checked the load profiling, then you load profiling, then you talk about talk about talk about purchasing new purchasing new purchasing new equipment and equipment and equipment and whether you will be able to whether you will be able to whether you will be able to use it use it use it effectively. You effectively. You effectively. You discuss the discuss the discuss the return on investment return on investment return on investment (ROI) with yourself, or maybe you (ROI) with yourself, or maybe you (ROI) with yourself, or maybe you just give in to just give in to just give in to the urge and buy the urge and buy the urge and buy it. I know I'm it. I know I'm it. I know I'm having this having this having this conversation with myself because these conversation with myself because these conversation with myself because these things are expensive. things are expensive. things are expensive. One thing I will say: when One thing I will say: when One thing I will say: when you're thinking about you're thinking about you're thinking about buying a new buying a new buying a new computer, if you computer, if you computer, if you 're buying a new one, 're buying a new one, 're buying a new one, leave the old one. Let leave the old one. Let leave the old one. Let it work. After all, if it work. After all, if it work. After all, if you can you can you can support its support its support its work and transfer work and transfer work and transfer it to agents, you can it to agents, you can it to agents, you can perform many perform many perform many slower tasks, slower tasks, slower tasks, offload offload offload calculations, and calculations, and calculations, and use use tools like Tailscale for this. tools like Tailscale for this. So, if you're So, if you're So, if you're starting your starting your starting your hardware search, let's hardware search, let's hardware search, let's first first talk about memory usage where agents memory usage where agents are today, before discussing brands are today, before discussing brands are today, before discussing brands . My . My . My guess, unless you are a guess, unless you are a super- super- heavy user, is heavy user, is to work with to work with 24–32 gigabytes of 24–32 gigabytes of 24–32 gigabytes of memory for now. 16 memory for now. 16 memory for now. 16 gigabytes may be gigabytes may be gigabytes may be enough. But you enough. But you enough. But you will have to will have to will have to work more actively with work more actively with work more actively with the settings. A the settings. A the settings. A larger volume gives you larger volume gives you larger volume gives you more space for more space for more space for regular programs, regular programs, regular programs, things you things you things you use, and use, and use, and agency agency agency tools that tools that tools that need to be open need to be open need to be open at the same time. It's not at the same time. It's not at the same time. It's not something I can say something I can say something I can say will be hyper- will be hyper- optimized for Codex optimized for Codex optimized for Codex or Claude, but that's or Claude, but that's not really how I buy hardware. I'm not really how I buy hardware. I'm buying it for myself and buying it for myself and buying it for myself and assume that over time I'll have assume that over time I'll have assume that over time I'll have other other other tasks. When it comes tasks. When it comes tasks. When it comes to

to to storage, I storage, I storage, I consider half a consider half a consider half a terabyte as a terabyte as a terabyte as a starting point, and a starting point, and a starting point, and a terabyte when terabyte when terabyte when working with large working with large working with large videos, projects, and videos, projects, and videos, projects, and files. If you files. If you files. If you create websites create websites create websites or generate a lot of or generate a lot of or generate a lot of videos, you should videos, you should videos, you should rely on your rely on your rely on your actual work actual work actual work habits to determine habits to determine how much storage you will how much storage you will how much storage you will need. And just need. And just need. And just to note, you might to note, you might to note, you might ask, "Why is Nate ask, "Why is Nate ask, "Why is Nate talking about Mac and not talking about Mac and not talking about Mac and not Windows?" I'll be Windows?" I'll be Windows?" I'll be frank, as I've frank, as I've frank, as I've said many times before: the said many times before: the said many times before: the AI ​​revolution on Mac is AI ​​revolution on Mac is six to eight months ahead of Windows. Windows computers are Windows computers are getting AI, but not getting AI, but not getting AI, but not as fast, partly as fast, partly as fast, partly because all because all because all the labs are running the labs are running the labs are running Macs, and partly because Macs, and partly because Macs, and partly because Mac chips are better Mac chips are better Mac chips are better optimized for AI in optimized for AI in optimized for AI in general. So when I general. So when I general. So when I talk about Mac, I'm talk about Mac, I'm talk about Mac, I'm actually giving you actually giving you actually giving you tips that can be tips that can be tips that can be applied to other applied to other applied to other environments. But the environments. But the environments. But the main problem main problem main problem that that that most of us face, and most of us face, and most of us face, and the computers that " the computers that " suffer" are suffer" are Mac users' devices, because Mac users' devices, because they actively they actively they actively rely on rely on rely on agents and give them agents and give them agents and give them a lot of work. That's a lot of work. That's a lot of work. That's why I'm focusing why I'm focusing why I'm focusing on Mac. As for the Mac: on Mac. As for the Mac: on Mac. As for the Mac: If you're buying a new If you're buying a new If you're buying a new device, Apple device, Apple device, Apple offers the M4 Mini for offers the M4 Mini for offers the M4 Mini for $899, and that base $899, and that base $899, and that base model has the same 16 model has the same 16 model has the same 16 gigabytes of memory gigabytes of memory gigabytes of memory and a quarter terabyte ( and a quarter terabyte ( 256 GB) of storage. This is 256 GB) of storage. This is 256 GB) of storage. This is a bit too small for a bit too small for a bit too small for the configuration I the configuration I the configuration I described, but it's described, but it's described, but it's workable. The M6 ​​Mini workable. The M6 ​​Mini is another option, is another option, configurable with up configurable with up to 32GB of storage to 32GB of storage to 32GB of storage , while the M5 Pro Mini starts , while the M5 Pro Mini starts , while the M5 Pro Mini starts at $1,699 and at $1,699 and at $1,699 and offers up to 64GB of offers up to 64GB of offers up to 64GB of storage, which is even more.

storage, which is even more. storage, which is even more. And about the Pro version, And about the Pro version, And about the Pro version, you know, you have to you know, you have to you know, you have to decide if you decide if you decide if you really need it, right really need it, right ? You're going to ? You're going to ? You're going to invest in invest in invest in more more more RAM RAM RAM and in stable and in stable and in stable local operation, local operation, local operation, probably with probably with probably with local models, local models, local models, and I want you to and I want you to and I want you to think about it carefully. I've think about it carefully. I've think about it carefully. I've made videos about Macs made videos about Macs made videos about Macs before, and I realized I before, and I realized I before, and I realized I was talking about them as was talking about them as was talking about them as objects to objects to objects to buy. I've talked buy. I've talked buy. I've talked about them as about them as about them as shopping products and a little bit shopping products and a little bit shopping products and a little bit about about about AI workloads, but the AI workloads, but the AI workloads, but the reason I'm reason I'm reason I'm recording this video recording this video recording this video now is because now is because now is because the more we the more we the more we rely on rely on rely on agents, the closer agents, the closer agents, the closer this this this relationship becomes. The relationship becomes. The relationship becomes. The more I understand what more I understand what more I understand what my agents do, my agents do, my agents do, the better my the better my the better my buying decision will be. buying decision will be. buying decision will be. So in this video we So in this video we So in this video we try to try to try to make sure we make sure we make sure we understand it well. understand it well. understand it well. So, if you're So, if you're So, if you're a traveler, a MacBook might a traveler, a MacBook might a traveler, a MacBook might be a logical be a logical be a logical choice, because of choice, because of choice, because of course you just course you just course you just throw it in your bag throw it in your bag throw it in your bag and you can go anywhere and you can go anywhere and you can go anywhere . I think if . I think if . I think if you're buying a MacBook, you you're buying a MacBook, you you're buying a MacBook, you want to make sure want to make sure want to make sure it won't get in your way it won't get in your way it won't get in your way on the road. And this is where on the road. And this is where on the road. And this is where Tailscale becomes a Tailscale becomes a Tailscale becomes a great remote great remote great remote way to monitor and way to monitor and way to monitor and control other control other control other machines, so you machines, so you machines, so you can can can use use use Tailscale to control a Mac Tailscale to control a Mac Tailscale to control a Mac mini or an old mini or an old mini or an old laptop while having your laptop while having your laptop while having your work work work device with you. But even then, device with you. But even then, device with you. But even then, I I I believe that the computer believe that the computer believe that the computer in front of me should be able to in front of me should be able to in front of me should be able to handle handle handle the tasks that are in the tasks that are in the tasks that are in front of me.

front of me. front of me. So make sure So make sure So make sure you have thought everything through carefully you have thought everything through carefully you have thought everything through carefully . . . Chipmakers, Chipmakers, Chipmakers, including Apple, are talking a lot including Apple, are talking a lot including Apple, are talking a lot about about about neural processors, neural processors, neural processors, or NPUs, these days. They talk or NPUs, these days. They talk or NPUs, these days. They talk about them as if they are the about them as if they are the about them as if they are the only answer to only answer to only answer to questions about AI. I'm questions about AI. I'm questions about AI. I'm here to tell you: here to tell you: here to tell you: you can you can you can use this use this use this for your work for your work for your work tasks. This is not a tasks. This is not a tasks. This is not a magic panacea. You are magic panacea. You are magic panacea. You are still buying still buying still buying equipment to equipment to equipment to do a job, and do a job, and do a job, and you have to consider whether you have to consider whether you have to consider whether it will be up to the it will be up to the it will be up to the task or not. And task or not. And task or not. And if you optimize the if you optimize the if you optimize the entire system—how entire system—how entire system—how the memory works, how the the memory works, how the the memory works, how the storage works, storage works, storage works, how the data processing works how the data processing works how the data processing works —you don't just —you don't just —you don't just optimize one optimize one optimize one magic chip and magic chip and magic chip and call it call it call it done. So don't done. So don't done. So don't let yourself be let yourself be let yourself be fooled by promises of fooled by promises of fooled by promises of magical functions. You magical functions. You magical functions. You need to really need to really need to really understand how it all understand how it all understand how it all works together, and the works together, and the works together, and the quality audit quality audit quality audit I keep I keep I keep reminding you of will help you reminding you of will help you reminding you of will help you with that and allow you to with that and allow you to with that and allow you to reasonably assess whether reasonably assess whether reasonably assess whether you are lacking in you are lacking in you are lacking in CPU power CPU power , memory, or , memory, or , memory, or storage capacity. If running storage capacity. If running storage capacity. If running models locally is models locally is models locally is one of your goals, one of your goals, one of your goals, and I know many and I know many and I know many of you of you of you are passionate about this, the amount are passionate about this, the amount are passionate about this, the amount of memory available of memory available of memory available to the accelerator and the to the accelerator and the software it software it supports will be supports will be supports will be key factors key factors key factors in configuring in configuring in configuring your system. If your system. If your system. If you have a Mac Studio with a lot of you have a Mac Studio with a lot of you have a Mac Studio with a lot of memory, an memory, an memory, an Nvidia system, or Nvidia system, or Nvidia system, or perhaps an AMD computer perhaps an AMD computer perhaps an AMD computer with unified with unified with unified memory, think memory, think memory, think about how these different about how these different about how these different memory architectures memory architectures memory architectures allow you to run allow you to run allow you to run different tasks. I would different tasks. I would different tasks. I would recommend doing a recommend doing a recommend doing a thorough analysis of thorough analysis of thorough analysis of your current your current your current system and agents, system and agents, system and agents, discussing in detail discussing in detail how your workloads how your workloads consume memory and consume memory and CPU cores, and CPU cores, and CPU cores, and then modeling this then modeling this then modeling this for different technical for different technical for different technical configurations to configurations to configurations to understand where you will understand where you will understand where you will get the best

get the best get the best bang for your bang for your bang for your buck. If you buck. If you buck. If you choose a Mac, the choose a Mac, the choose a Mac, the Apple Silicon-based device with its Apple Silicon-based device with its Apple Silicon-based device with its large large large memory capacity will allow memory capacity will allow memory capacity will allow you to run both you to run both you to run both local models and local models and local models and your own native your own native your own native Mac applications. If Mac applications. If Mac applications. If you work in the you work in the you work in the Nvidia ecosystem, the Nvidia ecosystem, the Nvidia ecosystem, the approach approach approach using Nvidia using Nvidia using Nvidia GeForce RTX is extremely GeForce RTX is extremely GeForce RTX is extremely powerful. It is powerful. It is powerful. It is actually designed actually designed actually designed to help to help to help run local run local run local models, but requires models, but requires models, but requires CUDA, a CUDA, a CUDA, a computing computing computing platform from Nvidia. And platform from Nvidia. And platform from Nvidia. And if your work if your work if your work depends on depends on depends on GPU tools and you're GPU tools and you're GPU tools and you're doing doing doing optimization or optimization or optimization or model training, model training, model training, that works great, that works great, that works great, right? If you're that kind of right? If you're that kind of tech fanatic, and I know tech fanatic, and I know many of you are, that's many of you are, that's great. Moving on great. Moving on from Nvidia: if you from Nvidia: if you from Nvidia: if you work in an work in an work in an AMD environment, although, AMD environment, although, AMD environment, although, according to observations, there are according to observations, there are according to observations, there are few such people, few such people, few such people, you should assemble you should assemble you should assemble a configuration with a a configuration with a a configuration with a large amount of large amount of large amount of memory so that a memory so that a memory so that a regular PC on Windows regular PC on Windows regular PC on Windows or Linux works or Linux works or Linux works stably. There are also stably. There are also stably. There are also people who people who people who use Omaki and use Omaki and use Omaki and that works great too. that works great too. that works great too. I released a whole video I released a whole video I released a whole video about Omaki and how about Omaki and how about Omaki and how this platform this platform this platform is designed is designed is designed specifically for specifically for specifically for agents. This is quite agents. This is quite agents. This is quite correct in this correct in this correct in this case as well. Nvidia Spark has its case as well. Nvidia Spark has its case as well. Nvidia Spark has its own software own software own software requirements. I could go on and on requirements. I could go on and on requirements. I could go on and on .

. . My goal here is to My goal here is to get you get you get you thinking about thinking about thinking about what level of what level of what level of maintenance maintenance maintenance you are willing you are willing you are willing to provide for your equipment and to provide for your equipment and to provide for your equipment and how deep how deep how deep you want to go into you want to go into you want to go into profiling. The profiling. The profiling. The more resources you have more resources you have , the less strict and , the less strict and , the less strict and relaxed you relaxed you relaxed you can be in can be in can be in profiling, and the more profiling, and the more profiling, and the more comfortable you will be comfortable you will be comfortable you will be using using using the computer while the the computer while the the computer while the agent or agents are agent or agents are agent or agents are also also also using it. using it. Overall, I would say Overall, I would say that the least technically that the least technically that the least technically difficult choice is Mac, difficult choice is Mac, difficult choice is Mac, and then the level of and then the level of and then the level of complexity only complexity only complexity only increases. And by the increases. And by the increases. And by the time you move time you move time you move to Nvidia or AMD, if you to Nvidia or AMD, if you to Nvidia or AMD, if you install an install an install an operating system operating system operating system like Ubuntu, you're a like Ubuntu, you're a developer, you're a developer, you're a developer, you're a technical person, and technical person, and technical person, and you're comfortable with that. You you're comfortable with that. You you're comfortable with that. You should be inspired by should be inspired by should be inspired by this, because you will this, because you will this, because you will gain much gain much gain much more control. It's more control. It's more control. It's like a manual like a manual like a manual transmission versus an transmission versus an transmission versus an automatic in a automatic in a automatic in a car. If you car. If you car. If you say, "Hey, I say, "Hey, I say, "Hey, I want to go deeper. I want to go deeper. I want to go deeper. I want want want to understand more deeply the tradeoffs to understand more deeply the tradeoffs to understand more deeply the tradeoffs when choosing local when choosing local when choosing local models." I have this in the models." I have this in the computer guide for Substack. An computer guide for Substack. An important difference important difference important difference is that is that is that the memory needed the memory needed the memory needed to accommodate the to accommodate the to accommodate the local model is an local model is an local model is an additional requirement. additional requirement. additional requirement. So think about So think about So think about your strategy for your strategy for your strategy for local models.

local models. local models. If you are trying If you are trying If you are trying to buy a machine with a to buy a machine with a to buy a machine with a lot of lot of lot of memory, memory, memory, make sure you are make sure you are make sure you are doing it for a good doing it for a good doing it for a good reason. Either you reason. Either you reason. Either you run many run many run many cloud models that cloud models that cloud models that use use use many local many local many local files and files and files and tools, or you tools, or you tools, or you run a large run a large run a large local model. local model. local model. Ultimately, you have three Ultimately, you have three Ultimately, you have three options, while options, while options, while most people most people only have two. You can only have two. You can only have two. You can keep the car keep the car keep the car you have, that's definitely an you have, that's definitely an you have, that's definitely an option that we all have option that we all have option that we all have . You may . You may . You may need to need to need to profile her more actively. I profile her more actively. I profile her more actively. I talked about this earlier talked about this earlier talked about this earlier in this video. In in this video. In in this video. In some cases, an some cases, an update can be made. In many update can be made. In many cases, they cases, they cases, they block this block this block this possibility and make possibility and make possibility and make it more difficult. Mac, it more difficult. Mac, it more difficult. Mac, of course, makes this more of course, makes this more of course, makes this more complicated. Or you complicated. Or you complicated. Or you can buy a new can buy a new can buy a new car and car and car and use the old one for use the old one for use the old one for agent tasks. All of agent tasks. All of agent tasks. All of these options are these options are these options are real. I just real. I just real. I just want you to make want you to make want you to make choices that are choices that are choices that are conscious and take into account the conscious and take into account the conscious and take into account the development of agents. development of agents. development of agents. So, the fourth So, the fourth So, the fourth principle, and I know principle, and I know principle, and I know you're wondering what this is all about you're wondering what this is all about you're wondering what this is all about . The fourth . The fourth . The fourth principle is principle is to make sure to make sure to make sure you have an ongoing you have an ongoing maintenance plan. maintenance plan. The machine you The machine you The machine you use use use interactively interactively interactively needs your attention needs your attention needs your attention when it is running with when it is running with when it is running with difficulty. The machine difficulty. The machine difficulty. The machine you leave you leave you leave running in the background running in the background running in the background with Tailscale may with Tailscale may with Tailscale may just be doing just be doing just be doing the job, and it has the job, and it has the job, and it has other other other maintenance and maintenance and maintenance and attention requirements. So when we attention requirements. So when we attention requirements. So when we talk about AI and talk about AI and talk about AI and how we manage how we manage how we manage it over time. I'm a it over time. I'm a it over time. I'm a big proponent of big proponent of big proponent of giving your giving your giving your agent a regular agent a regular agent a regular job of maintaining job of maintaining job of maintaining your your your computer's health. This, by computer's health. This, by computer's health. This, by the way, also applies to the the way, also applies to the the way, also applies to the agents we agents we agents we use on use on use on virtual machines.

virtual machines. virtual machines. I have an entire I have an entire I have an entire Grok Bot agent that Grok Bot agent that Grok Bot agent that specializes in specializes in specializes in maintaining the health of maintaining the health of maintaining the health of my Linux Grok Bot. This is a my Linux Grok Bot. This is a my Linux Grok Bot. This is a Linux installation, it Linux installation, it Linux installation, it needs to be working, and needs to be working, and needs to be working, and Grok Bot is Grok Bot is Grok Bot is obsessed with it. You obsessed with it. You obsessed with it. You can adopt this can adopt this can adopt this approach for your approach for your approach for your local machine. You local machine. You local machine. You can ask can ask can ask the agent to monitor the agent to monitor the agent to monitor folders so that it folders so that it folders so that it understands when something understands when something understands when something changes significantly and is changes significantly and is changes significantly and is worth paying worth paying worth paying attention to, to keep attention to, to keep attention to, to keep work active, work active, work active, prioritize prioritize prioritize human actions over human actions over human actions over agent actions, and move agent actions, and move agent actions, and move agent work to the agent work to the agent work to the background. Or you background. Or you background. Or you can change that can change that can change that if you want, but if you want, but if you want, but I like I like I like being able to being able to being able to do things do things do things on my own computer. on my own computer. on my own computer. And all of this can be done And all of this can be done And all of this can be done with a high level of with a high level of with a high level of confidence because confidence because confidence because these agents are native these agents are native these agents are native to computing. And that's to computing. And that's to computing. And that's one of the most one of the most one of the most important important important fundamental fundamental fundamental things that's changed things that's changed things that's changed in the last year, in the last year, in the last year, right? We have always had right? We have always had right? We have always had LLMs in the age of artificial LLMs in the age of artificial LLMs in the age of artificial intelligence, starting intelligence, starting intelligence, starting with ChatGPT. We didn't always with ChatGPT. We didn't always with ChatGPT. We didn't always have confidence that have confidence that have confidence that these agents were these agents were these agents were intelligent enough intelligent enough intelligent enough to effectively to effectively to effectively help us help us help us manage our manage our manage our computing computing computing workloads. workloads. workloads. Now we have. By the way, Now we have. By the way, Now we have. By the way, I've mentioned Tailscale a few times already I've mentioned Tailscale a few times already I've mentioned Tailscale a few times already . If . If . If you didn't understand what it you didn't understand what it you didn't understand what it is, it's a way to is, it's a way to is, it's a way to control machines control machines control machines over a private over a private over a private network. In other network. In other network. In other words, Tailscale words, Tailscale words, Tailscale provides a private provides a private provides a private connection connection anywhere in the world, anywhere in the world, anywhere in the world, allowing you allowing you allowing you to work on your to work on your to work on your computer computer computer remotely and remotely and remotely and securely. So many securely. So many securely. So many of us of us of us use Tailscale with a use Tailscale with a use Tailscale with a local machine local machine local machine at home, and wherever we at home, and wherever we at home, and wherever we are, we can are, we can are, we can give that machine give that machine give that machine instructions for the instructions for the instructions for the agent, and it's really agent, and it's really agent, and it's really very easy to very easy to very easy to manage. It's not manage. It's not manage. It's not too difficult too difficult too difficult to set up. One to set up. One to set up. One way you can way you can way you can save hundreds of save hundreds of save hundreds of thousands of dollars in thousands of dollars in thousands of dollars in computing is to

computing is to computing is to remember that you remember that you remember that you probably have an probably have an probably have an old computer lying around, an old computer lying around, an old computer lying around, an old Mac mini, an old old Mac mini, an old old Mac mini, an old Windows laptop, or an Windows laptop, or an Windows laptop, or an old Mac, whatever. And old Mac, whatever. And old Mac, whatever. And you can say, "I'll you can say, "I'll you can say, "I'll plug it in." "I'll plug it in." "I'll plug it in." "I'll use Tailscale to use Tailscale to use Tailscale to control it, make control it, make control it, make it a computer for the it a computer for the it a computer for the agent, and the agent can agent, and the agent can work there independently and work there independently and perform tasks perform tasks perform tasks for me." This is much for me." This is much for me." This is much cheaper than buying a cheaper than buying a cheaper than buying a new one. And if you're a new one. And if you're a new one. And if you're a tech geek, Raspberry Pi has tech geek, Raspberry Pi has tech geek, Raspberry Pi has been around for quite some been around for quite some been around for quite some time. These are small time. These are small time. These are small boards, and they boards, and they boards, and they play a similar but play a similar but play a similar but smaller role in this smaller role in this smaller role in this world. You can world. You can world. You can use a use a use a Raspberry Pi to control Raspberry Pi to control e-ink displays . I definitely do this . I definitely do this . I definitely do this at home, and you can at home, and you can at home, and you can use them use them use them to run to run to run small local small local small local services while your services while your services while your laptop is somewhere else laptop is somewhere else laptop is somewhere else . And again, . And again, . And again, this is where Tailscale can come in this is where Tailscale can come in this is where Tailscale can come in handy. handy. handy. Ultimately, it's your Ultimately, it's your Ultimately, it's your decision how much decision how much decision how much computing computing computing power you power you power you want to allocate. want to allocate. want to allocate. Part of the reason I'm Part of the reason I'm Part of the reason I'm making this video is because I making this video is because I making this video is because I want to save you want to save you want to save you money. If you money. If you money. If you use your use your use your computer to its fullest, computer to its fullest, computer to its fullest, you are less likely to need to you are less likely to need to you are less likely to need to buy a new one. And buy a new one. And buy a new one. And if you do buy, if you do buy, if you do buy, you'll do it you'll do it you'll do it smarter. So, smarter. So, smarter. So, think about it this way. think about it this way. think about it this way. We're trying to We're trying to We're trying to figure out how to figure out how to figure out how to get the most out get the most out get the most out of co- of co- of co- living with our living with our living with our agents today. We agents today. We agents today. We have the house we have the house we have the house we have. We have the have. We have the have. We have the rooms we have. We rooms we have. We rooms we have. We try to try to try to make sure, before make sure, before make sure, before we buy a new we buy a new we buy a new house, metaphorically house, metaphorically house, metaphorically speaking, before speaking, before speaking, before we buy a new we buy a new we buy a new computer, that we're computer, that we're computer, that we're making the making the making the most of what we most of what we most of what we already have. And if already have. And if already have. And if we we we have an old abandoned have an old abandoned have an old abandoned shed in our backyard, which is my shed in our backyard, which is my shed in our backyard, which is my favorite metaphor favorite metaphor favorite metaphor for an old for an old for an old computer somewhere computer somewhere computer somewhere under your feet, we'll under your feet, we'll under your feet, we'll use it. We'll use it. We'll use it. We'll make sure we have it make sure we have it make sure we have it and that it's and that it's and that it's part of our part of our part of our living space, living space, living space, part of our part of our agent computing space, agent computing space, before we run out to before we run out to before we run out to buy more. Because I know buy more. Because I know we all love we all love we all love unwrapping presents unwrapping presents unwrapping presents under the tree, but I under the tree, but I under the tree, but I also like to

also like to also like to make sure I'm not make sure I'm not make sure I'm not wasting my wasting my wasting my money. So I want to money. So I want to money. So I want to make sure make sure make sure you have the tools you you have the tools you you have the tools you need to need to need to make an informed make an informed make an informed purchasing decision, purchasing decision, purchasing decision, as well as to as well as to as well as to confidently confidently confidently use what use what use what you already have. you already have. you already have. OK. So, let's OK. So, let's OK. So, let's identify the key identify the key identify the key points you points you points you can work with. can work with. can work with. Of course, if you Of course, if you Of course, if you want deeper want deeper want deeper details, it's details, it's details, it's much deeper than much deeper than much deeper than I can cover in the I can cover in the I can cover in the video. Go to the video. Go to the video. Go to the guide on Substack. You guide on Substack. You guide on Substack. You can literally can literally can literally feed it to feed it to feed it to your agent and they your agent and they your agent and they can work with it can work with it . It will cover all the . It will cover all the . It will cover all the different tools different tools different tools I talked about. It will I talked about. It will I talked about. It will also cover the audit also cover the audit also cover the audit described in the manual. described in the manual. described in the manual. This will be very useful This will be very useful This will be very useful if you are trying to if you are trying to if you are trying to efficiently squeeze efficiently squeeze efficiently squeeze more out of the more out of the more out of the computing computing computing resources you have. resources you have. resources you have. Additionally, if your Additionally, if your Additionally, if your computer is running computer is running computer is running slow like mine, slow like mine, slow like mine, run run run diagnostics and diagnostics and diagnostics and make sure you make sure you make sure you understand your understand your understand your memory memory memory and storage usage. and storage usage. and storage usage. Make sure you Make sure you Make sure you understand understand understand the load on the load on the load on your processor cores. your processor cores. your processor cores. Make sure you Make sure you Make sure you understand what exactly is understand what exactly is understand what exactly is causing it and causing it and causing it and know how know how know how to profile your to profile your to profile your work. Profiling work. Profiling is a fancy word is a fancy word is a fancy word for for for sequencing work sequencing work so that you don't create a so that you don't create a so that you don't create a bunch of load bunch of load bunch of load at once when it at once when it at once when it peaks and just peaks and just peaks and just shuts down the machine. shuts down the machine. shuts down the machine. I had a memory leak last night I had a memory leak last night because, as a because, as a human, I opened a human, I opened a human, I opened a new program, and new program, and new program, and everything was fine before that.

everything was fine before that. everything was fine before that. This happens, but it This happens, but it This happens, but it happens less often happens less often happens less often if we take good care if we take good care if we take good care not to not to not to overload the overload the overload the system. But in general, system. But in general, system. But in general, we want to we want to we want to manage our manage our manage our demand intelligently so that we can demand intelligently so that we can demand intelligently so that we can handle handle handle concurrent requests. concurrent requests. concurrent requests. So the next time So the next time So the next time you sit down at your you sit down at your you sit down at your computer, I want computer, I want computer, I want you to think about it this way: you to think about it this way: you to think about it this way: I have neighbors. I have neighbors. I have neighbors. I have many agent I have many agent neighbors, and we have to neighbors, and we have to neighbors, and we have to agree on the agree on the agree on the rules of rules of rules of living together. We living together. We living together. We have to agree on have to agree on have to agree on who takes the place. We who takes the place. We who takes the place. We need to understand how we need to understand how we need to understand how we all all all use it to use it to use it to be productive. be productive. be productive. Essentially, it's a question of Essentially, it's a question of Essentially, it's a question of who does the dishes who does the dishes who does the dishes for the AI ​​and for you on for the AI ​​and for you on for the AI ​​and for you on your computer. your computer. your computer. Who takes care of cleanliness? No Who takes care of cleanliness? No Who takes care of cleanliness? No one wants to one wants to one wants to talk about washing talk about washing talk about washing dishes when you live with dishes when you live with dishes when you live with neighbors. This is the boring neighbors. This is the boring neighbors. This is the boring part, but it is part, but it is part, but it is very important to very important to very important to keep the keep the keep the house running. My goal in this house running. My goal in this house running. My goal in this video is to make sure video is to make sure video is to make sure you understand the you understand the you understand the basic elements: basic elements: basic elements: memory, processor, memory, processor, memory, processor, storage. You understand storage. You understand how how agents use them, how agents use them, how you use them, you use them, you use them, and ultimately have a plan for and ultimately have a plan for how to use how to use how to use them better or make a them better or make a them better or make a smart purchase for the smart purchase for the smart purchase for the future. Good luck future. Good luck future. Good luck and keep going, Hawks.