Build Your Own Roofing AI: Turn Roof Inspection Photos Into Scopes (No Code - Complete Guide)

All right, listen up, roofers. Today I'm going to show you how guys with no technical background whatsoever can build their own AI software. Train straight on your inspection process, starting from absolute zero. In this video, you're going to watch us build an AI system from scratch that turns those inspection photos into real scopes of work step by step, no coding experience required. So the dream would be that like in Zooper, you would be able to do an inspection in accordance with the process that we build in Zoober. And then you'd be able to build an inspection report with full size photos and you would download that inspection report, upload it to a new folder in your Wind Surf folder, and be like, I've created a New folder for 43 Brown street, please write me a scope. And it would just go through and follow the directives and executions and write you a scope. That's like my goal for this when it's done. But in the meantime you should just be able to upload a bunch of pictures and talk to it. But we're kind of gradually moving in that direction. Like to really create like this level of detail in SOPs and training documentation and to like communicate this to a developer and then walk away with like a built out environment for a piece of software like this would have been like a hundred grand three years ago. Download Windsurf. Click Next for VS code. You can click through and pick a color if you like dark, if you like light, whatever you like. And then, yep, sign up, sign up for a free trial. It'll burn up your credits free fairly quick. And then you'll eventually go to the signing up for Pro. You should have an email with three files in it that I sent you. Because then what I want you to do is I want you to put those into a folder. You can call it Go Roof365. And then you're gonna put those three files in there. When it says open folder, go in here and click Open folder and go find that folder. There you go. See the three files in there? So you can click Yep, trust the authors up here. If you click up here, this will open up the chat. Yep. And then down here where the model and then Change it to Gemini 3 Pro High. The reason that I picked Gemini was because it has a big context window, meaning that you can talk to it a lot. Before your context window is full, first thing you do, I want to type this out. Please instantiate my environment in accordance with the Gemini MD file. When you say Please instantiate my environment in accordance with the Gemini MD file. It doesn't matter whether you say Gemini MD or Agents MD or Claude md. The point is, whatever you see in the bottom right corner is the model that you're using. If it's OpenAI model, like a GPT, then you would say agents MD. If it's a opus or sonnet or Claude, anything that's by anthropic, then you'd say Claude md. If it was Gemini, then you would say Gemini md. And you only have to do this once. But if you switch models, then you might have to tell it, hey, read the Agents MD file so that it understands how its behavior is supposed to be. It's kind of like a behavior setting file. So you copied that out of the. There we go. Click run. Because, yeah, it's going to install like Python and stuff like that on your computer. It's basically going to start putting a lot of stuff into your libraries. And so you'll hit Accept already here. So think of like Python as like a bunch of code. And every time you, you want to like make Python do something, you go to the library, look up the function, and then you like add that function to your code just like it sounds. It's a library. Full instructions on how to write Python and make it work. So what you're doing is you're saying, hey, I want to download the library. Because the agent will just go read the Python library itself. So you can click run there and I'll kind of show you what I mean. So this is like extensions, right? And so, see, so it's going to install like Python and Python Debugger and all this kind of stuff. Over time, it's going to ask to install these different libraries. What this is called is an interactive developer environment. The environment just means a folder with a bunch of stuff inside. That's an environment. And then developer obvious in the center here, right? You'll see that it'll keep creating files. You have like a dot ENV file, you have a dot GITIGNORE file. If you go back to click on this above here, this little file files thing right here in the top left. If you click like Gemini md, it's the exact same as Claude MD and Agents md, it's the exact same. But this is where you view the files in your folder. When it creates a new Python script, you'll be able to see it. It'll be a tab up here. When you close this, like if you hit the X right there, you're not doing anything. It's not deleting it it's not going away, but sometimes if you change it, you'll have to save it. So go ahead and X this one out. But now just type Test API key up there. So the reason it creates lines, like whenever you hear like, oh, I wrote 40, 000 lines of code. Well, this was how like a developer knew that they would lines of code into this, which is the file viewer in the environment. But notice how this turned into a circle. So that means that this file has not saved. This over here, your agent that orchestrates it doesn't know that test API key has been typed yet because it hasn't been saved. It doesn't look here. This is like an editor. It looks at the file. So you click this circle up on the top right here. When you click that, it'll prompt you to save it. And you can go ahead and save it. It's not going to hurt anything. Now you hit save, it would know that that file says Test API key. Put your mouse around here and you should be able to drag this that way because that's where like the bulk of your time where your eyes are looking at stuff is going to be looking at this. This is like your ChatGPT. If you look down here, this little circle, if you click here over that. So you've used 2% of your context window. So now you can see how much of your context you've actually used. So what we want to do now is we want to kind of tell it what is the goal of this thing, is that to tell it that we want to be able to upload pictures to this thing and we want to be able to teach it how we estimate and we want to give it access to a vectorized knowledge base. I'm going to give you another prompt to give it here. I'm just going to use my voice to text to write it. The purpose of this environment and how I want you to consider directives and execution in accordance with the Gemini MD file is that you're going to help me build scopes and estimates for commercial roofing. For this, you will need a knowledge base. I have a vectorizer account and I have a vectorized store with a reg database that I am putting the NRCA National Roofing Contractors association documents in that will have a lot of the scope requirements and building codes and other things that you will need to get access to so that you can understand my industry. I am not a senior developer, so I will need your guidance. But I want to make sure that when we run into errors, you can Postulate ideas on how we can make it better. How I see this ideally working is I send you pictures that I can put into folders and you would create projects and then you would be able to find a way to analyze those pictures. And then I can give you inspection details or information that we collected while we were physically on site at a building. And then I would give you a request and you would help me build out a scope that I could hand off to my team so that they could turn that into a bid or an estimate for my customers. So I think you're going to need to connect to Vectorize through an API or an mcp and I would like you to guide me through that process. These could live as temporary files, but I want you to build our directives and our executions with Python script. And I think you will need to get the Python libraries, but if there are more libraries that we need or any other ideas that you have as to how we would accomplish this, I'm open to your suggestions. So you don't always have to go to this level. I'm just trying to get this environment set up really well. To start, you won't break it in an unbreakable way if once you like take over from here. I'm just trying to build on rock instead of sand so that when you then play with this, I know at least the foundation creation was built right? So I'm going to send you this. That was actually one of my questions. What happens if I break it? If you break it, we can use it to fix itself. And I'll actually go one step further once we're done this prompt, we'll do one more prompt and then I'll get you to duplicate this folder on your computer so that you could reopen this environment from this point. Kind of the same thing as like a GitHub because like we will, we will basically back up this environment the way that we've set it up. I shared you a document it when you're in here, there's two modes, see right here where it says code. So you can't do anything while it's running, but when it's not running you can click that and you can chat with it. So what that means is it will not edit anything. So click the drop down. I. I usually set this up. Always allow all requests. Yeah, allow all requests. Yeah. Same thing with the other one. You'll still have to accept all when it edits a file, but we're going to make it so you have to accept a lot Less. We're basically going to go into like full auto mode where this thing is almost agentic. So go ahead, go run. Yep, go run and go accept all. It's a lot of clicking like this at first. Go ahead and install right here. That's like I said, it's going to install Python Trust Publisher. It's still running. You can see here the it hasn't done a play button. So you're just basically watching what it does. Oh, scroll down. So it's getting run. Python's going to basically find out what version it is. So see this is like its to do list, right? And so it's going to fix itself. It's kind of doing all this work for you and it's working through its tasks. When it says those like dash dash like this, where it says like Python dash dash version, that's it, like running it in the terminal. As soon as this prompt is done running and we're gonna have to do a lot of this clicking, I'm going to turn on the automatic. Okay, perfect. So yeah, it's installing Python, so then might be saying, oh, trying to check to see if Python's installed and it's not installed yet. Okay, so close and reopen the application to refresh your system path. Okay, so before we do that, so what I want you to do is, is I'm going to say once your environment's orchestrator purpose is set up, this is a good opportunity to back up your original orchestration environment. Just so you could always come back to like square one. Please add a orchestrator purpose. MD file is similar to a readme file. That explains what your interpretation is of the orchestrator purpose and what you've set up so far so that it isn't dependent on the cascade memory. And I could duplicate this environment over and over and over again to try different setups in the future. Use this prompt to do that. Go accept file. So that's basically saying we created a file and now you can accept the file if you open up that file explorer again up here. So now it's created an orchestrator purpose MD file. So you can always say, hey, go look at your orchestrator purpose and it'll like understand what that is. Now you can click run. It's probably just going to check that we didn't restart it. So yeah, it's. We have to still restart Windsurf, but we've created our folder now. Oh yeah, because you know what, it's probably trying to execute a terminal in the Google Drive, which is why it's not maybe not working. You can close Windsurf now. Now go back to like your Windows File Explorer and I want you to duplicate this environment onto your C drive so you can just copy all that. So you've kind of got a backup of your environment in your Google Drive, but this will be the one that you actually work from because once you install Python, once, it'll work for all of your agents. Go into Windsert, reopen Windso, Surf again, and then you're going to open up this folder instead. Open up your Cascade, which is like your chat window. It's this one right here. And so then now what I would do is say please review the Gemini MD file and the Orchestrator underscore purpose MD file so that you understand this environment, how it's set up and your orchestration purpose and what we're trying to set up today. Yep. So it understands it. So I would say verify that you have installed Python, have the version that you need, and I want your help in setting this up. I think we're going to need to connect our Vectorize rag database. Next is probably the smartest first step is to make it so you understand the roofing industry, so you understand building code and building assemblies and roof products and diagnosing leaks, etc. This is error. Okay, so what we're going to do, go up into your settings. Auto execution. Click here and go Turbo. Okay. And then go here. Click Auto. Continue parallel searches as a sub agent. Yeah. Click Turn this on. Explain and fix. Yep. Click here. Yeah, Click here. Show notifications when a cascade finishes while in the background. Yep. Turn this on. Click on Manage rules. Show inline shortcut actions when the cursor is on an empty line. Yep. Scroll to Next diff on accept. Yep. Okay, so that's how you do this. You can hit Control W and it'll close the window you're in. That chat window is called Cascade. So you can open that up again. All right, so I've successfully verified Python environment, installed necessary dependencies. Okay. Connect your vectorize any day access to the code. All right, so I'm going to show you what I'm up to here. So I'm going to steal your screen share. All right, so we're going to go to Vectorize. There it is. There we go. So if I wanted to create an internal knowledge base of documents for the company, a tool like this be now it comes down to like, how would you access it? Right. So you'll probably use like HubSpot has a tool like this and then it's accessible within HubSpot. So you'd probably use HubSpot's knowledge base because that's like where more employees are going to get access to it. But yeah, if you put everything here, then you could literally connect a chatbot to it. So if you're looking in your window, it probably has something like I need the endpoint, the organization id, the pipeline id and a token. Okay, what I need you to do, you're going to give this to the chat, everything above the token. So then go into the env file, paste all that in there in that env file. Now copy where it said vectorize API key first, where vectorize API key equals copy that and then paste that on line nine. Now go copy the key and paste that in right after. Okay, so then you have to close it because it hasn't been saved yet. So now put something in the chat for you to write. Next, I have added the API key organization ID pipeline ID to ENV file as well as the. The endpoint is all saved in the dot ENV file. Please confirm. Okay, so now what we need to do is we need to go give it a access to an API key. Now copy that. Now go back to Windsurf and go to your env file again. Now you can go right here, it says LLM API key. I would actually change that to say OpenAI and then go ahead and paste it and then close it and save it and then hit Accept all here because it's trying to save a file. Now it's basically like creating some folders. Click Accept all here. The other thing is it, see, it says LLM API key. So we're gonna have to tell them like, hey, we actually saved it as OpenAI API key because it's going to go look for LLM API key, hit Accept all. That's the one annoying thing is you kind of got to sit here and babysit it and hit accept all so it can save files. Okay, so while that's working, go up here to the little plugin, little like electrical cord plugin looking thing. Okay. And go mcps. Click little settings. Yeah, I'm going to copy something into your, into your document that I want you to put in there. And then you go paste it into that. Okay. And then close it and save it. All right, so now click on your little plugin again. All right. And it's connected. Perfect. So you don't have to do anything in there yet. Once we get done, we're going to tell it, hey, you have access to this mcp, we'll end up saying that as well. Yeah, and our CA repair doc. Okay, perfect. So I got the right one. Okay, so I have successfully set the environment connected to the vectorized reg database and currently to the Apex roofing pipeline. Okay, so we don't want that. The NRCA pipeline is found to be in a shutdown state and returning errors. I switched to the function Apex roofing pipeline to ensure immediate usability. Okay, so we don't want that. So I'm going to say. I'm going to give you a prompt to say back. We have added an MCP server to the MCP config file for our NRCA repair doc. I want to make sure you install a guardrail to prevent using any other pipeline than NRCA doc, because that's the only one that has the proper scoping instructions. Anything else in those other pipelines would create false positives or incorrect answers, so don't use those. Now, I've also noticed that you have created a bunch of Python files in the main directory. What I want you to do is clean up the Python files you no longer need. You can move them to temp or you can move them to like a new folder if you want. But I want you to move the ones you're going to keep into the execution folders, if those are considered executions. After you clean this up, I think I want to first verify that you have access to the NRCA repair doc reg pipeline because it's now finished uploading all the files and should no longer be in shutdown status. And once you've done that, I will start to work with you on directives, so you can just copy that and paste it in. Then we're going to switch to chat mode just so you can see how that works. And then where it says code right here, change that to ask. So this just means that it can't edit anything. It's going to kind of talk to it. So click. Go ahead and click Send. So now it's not going to edit anything. It'll basically put together a plan and then you can review the plan. Use the ask mode often. Like when you're not with me, use the ask mode fairly often. You know, say, can you explain that plan? I don't understand. Can you explain that? Explain that to me like I'm five. Then you switch to code mode, execute the plan, and then gradually you get more and more comfortable just working in the code mode. Okay, so it's going to clean up the workspace, create a dot temp debug scripts directory. Move all the temporary Python files there. Restore and secure configuration. So edit the env file to switch the victorized pipeline back to the NRCA repair doc id. So now switch to code mode and then just say yes, proceed with step one. If you click up here on your little profile. Wind surf usage. So there you go. You've used 20 credits so far. You can close it. I was just curious to see where you're at. Okay, so I completed the cleanup, which it didn't do. Pipelines Interstate is active and populated. So now what we'll say is I'm going to give you a prompt here in the chat. I don't know why, but it doesn't appear that you did in fact move the temporary debug and test scripts into the temporary folder. And that was causing you quite a lot of extra credit use and errors. So I think you need to try again to move Python scripts to the proper temporary folder before we can start on the next steps. Throw that in the chat and see what happens in code or chat. No code is fine. There we go. I like to usually keep the directives folder expanded and the execution folder expanded just because I can see what directives it's building. Okay, remember you sent me a file that was like the Go roof inspection process? Yeah. Because what I'm thinking in my head, what we've done is we've given this thing access to a bunch of files, but we haven't really told it how we inspect. So what I would do is I would want to drag that inspection document, open it up in Google Docs, and I want to kind of download it as what's called a markdown file. Notice how all of these are like MD files. Okay, so click Download as markdown. Now open up a new chatgpt tab. So now upload that markdown file here. So click the plus and then add photos. Okay, so open it, go up here. Click Change this to pro. All right, now I'm going to give you a prompt to put in the chat. I have attached our inspection SOP as it currently exists. It's not fully complete yet, but that's okay. What I would like you to do is build me a training document for my AI agent that explains our inspection process, why we have each step on our inspection process, and what common things there are look for. You have seven files in your knowledge base that teach you all about the different types of roofing materials, etc. So you have been an effective tool for building scope. But I'm building a new agent that operates on OpenAI models, and I've put all of my files into a vector store. But what I want to give it is kind of a training document to understand our inspection process. So I need you to sequentially go through this entire inspection SOP and write detailed instructions for the why, the what, the how for each question, and create a markdown file that will be something I can download and deliver to my agent as a training document so it understands our inspection process in full. That way, when I give it inspections, it understands why many of these questions are there, so it understands what to do with photos, etc. All right, there we go. We got the prompt. Yeah. Yep. And you can send that over. It's been working on this thing now for 12 minutes. The prompt we gave it, basically the nice thing about it is it, like, it really forced it to, like, use its brain. All right, here we go. Let's see. All right, so scroll all the way to the bottom. Okay, so hit copy, open up a new tab and type docs new. Now paste. Okay, and then title this inspection SOP AI Training, and then download as markdown and put it in that same folder. So let's go back to Windsearch. All right, now bear with me one second. I'm going to create a new prompt. I've uploaded two files to the folder. One is an inspection SOP and the other one is an inspection SOP AI training document. This will help us with guiding the creation of your directives and executions. I need you to start building a knowledge base folder that allows you to understand understand how we do inspections of roofs so you can start to understand what types of photos I'm going to give you to analyze. So the idea is we're going to go out and do inspections. They might not always follow this entire process entirely just because it's a service work, meaning repairs and maintenance type quotes that we're going to be providing customers. But this tells you at least what the inspection is going to look like coming in. And then we have to build directives for how we're going to output scopes of work. I still have to train you on that. But first I need you to analyze those two documents in an intelligent way, using a high amount of reasoning to build a knowledge base that you would understand. And I think that will help you refine the Orchestrator purpose file, because you will understand a lot more about what type of inputs you can be expecting. So once you've done the review, I would like you to create a knowledge base folder and start adding Your own markdown files that are pieces of information you might reference when creating new directives and new executions. Okay, so now I want to optimize that prompt, and then we're going to give it to your Windserve. So while that's going, do you have any documentation about what a scope is supposed to look like? Kind of like the how to write a scope. Like, when you were in ChatGPT, you showed instructions. Like, remember how you said you've told it to? Like, numbers and bullets? Like, where did you teach it that? I just wrote it into the instructions. Okay, so go into ChatGPT, copy those instructions out of your service work folder and put that into a new markdown file and go get your instructions. Copy all this and then do docs new again. Okay, Go up to the top there and just say, like, GBB means good, better, best. Okay, so you have a new prompt to give your windsurf. I'm going to start working on the next prompt here, the word. Am I placing this in windsurf? Yeah, with code or. Yeah, with code. Yeah. All right, now we're going to give another prompt to Chat GPT from that new markdown file. So that file was called like instructions. I have given you a markdown file that was my instructions from my folder in my Chat GPT account. I want to convert this into agent directives for my idea agent that I've built. What I need you to do is review this markdown folder, analyze it for the the what, the how, and the why, and improve it so that I can give this to my agent as a prompt for it to review the file that you output. And the file that you output will be the instructions that are the input for my agent to understand how we design or write scopes. Because the agent's purpose, when done, will be to take inspections or photos that I give it and write scopes based on the information provided. But it needs to understand how to write a scope, the formatting, the language. It also needs to understand that it has access to a knowledge base in vectorize. Previously, in the ChatGPT folder, I had the NRCA documents in the folder. But now that we're moving this to an agent, it exists as a reg vector database that it has access to via MCP and API. So we need to kind of reformat these instructions so that it understands the source of its knowledge will be a knowledge base and the RAG vector database, and that will be where it understands the specific technical details of a scope. But once it understands the problems with the roof from the inspection and the photos. And it. It knows how to review or where to go to review its need for technical specifications, which is the knowledge base in the Vector store. Then it uses these instructions that you have reformatted to write the scope according to our company's design. So the action I want you to take is to review this document and reformat it in no less than 3,000 words as the instructions for how we write scopes and how it is to function in inside the environment, so that it understands what its Orchestrator Purpose document should contain, the directives it should create, which are the markdown files of sops goals and desired outcomes, and then what type of Python scripts it should create in its Executions folder to create controlled outputs. That really stretched me. So go back to Windserve. I just want to see what our progress has been like, where it started responding. Okay, so it created knowledge base, created voter standards, drainage, penetrations. Perfect. Keep scroll down. Okay. Okay, great. So now save that new markdown file that you were creating in Google Docs, because now we are finally at the place where we're about to give it its directives and executions. Like it's going to now understand what its purpose is after you give ChatGPT Pro this prompt that I'm about to give you and the file that you're about to download. Yep, that's perfect. Name. Yep. So download that as a markdown. Maybe don't save this to your Wind Surf file. Don't save this one. You can just put it anywhere. But just don't put it in the Windser folder because I don't want it to conflict with Orchestrator purpose. Right, Great. All right, now go back to your chat GPT that we were in that Pro1 that we already did. Yeah. So now upload the new markdown file that we just saved and I'm adding a new prompt. Okay. Yep, you can copy that in. All right, so this is exciting because this is going to be the file that we deliver it to. Say, like, so I guess I should start writing that prompt because we're going to get an output from this, build out our directives and executions based on the training document, provided context previously. The job that I'm asking for you to help me with was performed inside of a ChatGPT chat window, which means I did not have control over the memories. At best, what I had was using the Chat GPT feature of a folder, and in that folder I had the NRCA documents that you now have access to via the vectorize API and mcp And I couldn't really edit the knowledge base. And when I would upload photos or inspections or other source material, I would prompt it to write a scope. And because I did not have controlled repeated prompts or access to the constantly evolving memories or the ability to select specific parts of the NRCA documents, we would frequently run out of context and eventually get more hallucinations. So that led to me creating you and our file structure that we've been building. So what I did was I asked Chachi PT to build me a training document for you as my Orchestrator, to try and give you as much of a head start on the knowledge that it had learned from working with me, and to write detailed instructions that are contained in the document that I have provided. Just so you know, we're gonna have to tell it which docum document this is, because we don't have it yet. So this document is to be treated like your initial training for your Orchestrator purpose and to help you understand what directives and executions you should start creating, please review this document in full and come up with a plan for directives that you can create and executions that you should write for. Accessing your knowledge base. Accessing your vectorize RAG database in such a way that is aligned with the Gemini MD file and your Orchestrator purpose MD file so that you can start to build this environment to be a seasoned veteran scope writer that takes the inputs that I provide you in the Cascade Chat, which will be pictures and inspection reports, and for you to make your best attempts at writing a scope. When you write those scopes, I will likely give you feedback and you will add those things to your knowledge base so that you can self anneal and get better over time. Time please output a plan and start building directives and executions to get a preliminary environment set up that we can start testing. All right, so I'm optimizing that prompt. And now it's optimized. Okay. Yeah. So when this output comes out from ChatGPT, you're going to copy the output from this, you're going to go docs new and you're going to paste that into document, give it a quick review, save it as a new markdown file in your Windsurf folder, and remember the name that you named it. You're gonna add that file name into the prompt in the final training tab of the GoRuf365AI document. I saved those instructions. Okay. Essentially, I said that my goal was to make it so that Zooper you could download inspection report. But like the next Level is that Zooper has an API and it has an mcp. Like that thing that we connected, that MCP thing. Zooper has all that. So just like how we connected the AP and the MCP to Vectorize in theory, right? And again, this is all the fuzzy image is that you might just be able to go into that wind surf and be like, hey, my inspector just finished an inspection at 43 Brown Street. It's project 123456. Can you write my scope? And it will just go pull the inspection and pull the photos. And you wouldn't have to like download it, upload it, you know, like, you would have to go through that step. And that's why I said the future of work will be a lot of like, like the black screen with white text because the field workers will go to use these functional tools. But then you will be able to like have this back end thing because then the next step is like we create a custom property or field inside Zooper, which is like, right Scope, and then we take this agent that you're right now building locally and we upload it to an environment like they call it Docker. And then now you'd have the ability to essentially say, when we flip this switch in super go trigger the directive in the Docker hosted environment that triggers the directive, that then runs the executions, that then writes the scope. And then we connect Google Docs to the environment and it writes the scope into a Google Doc that you can then have upload to a file in HubSpot that you can then just attach to an email. It's not that far away. We just have to like, like brute force our way through the middle part. The making the external parts enact this agent isn't hard. It's just the training of the agent that's going to take time. Okay, so click Download the full directive package. Save it here, because we have to unpack it first. What's inside? Okay, so now scroll down. Man, this is amazing. Like, it's already written what it thinks might work. Okay, so go back to your folder because we're going to have to rename one of those files because it renamed one it called one Orchestrator Purpose. So we have to like go into that folder. Okay, so now rename Orchestrator Purpose. Rename it like Orchestrator Training. Change the name of the executions folder to like, just add examples to the end of it. And then SOPs change to SOP's examples and then go up one level because we're going to rename the main folder Rename that to like ID Agent Training. So change directives to Training Training. Now drag that into your windserve folder. Okay, so now I'm going to edit this prompt that we gave you. So provided training documentation in the IdeAgent training folder, which includes an Orchestrator training document, which is your starting point. But it also has some example SOPs and execution examples. You don't necessarily want to use these before doing your review, but it should be a good starting point point. Okay, copy everything from developer down. Okay, so now you're going to paste that in there. All right. And you've got your folder in there. Yeah. ID Agent Training. Scroll up a bit. Go where it says afterwards says Starting point. Just click where it says after it says starting point right here. And go space at sign, type ide. Yeah, and then click ID Agent Training right here. Yeah, that's just making sure it knows where to go look and then let her go. You're gonna have to accept hit accept all quite a bit now. Yeah, we're just gonna train this agent up and then the next step is just to open up. The final thing is we're just gonna say like, okay, I think you've probably got us into a good place where you have built a good orchestration plan. I want to be able to close this chat cascade window, but before we do that, I want to make sure you haven't committed anything to your memories that should be be saved into either the knowledge base or orchestration files or execution files. I just want to make sure there's nothing that we're going to lose in terms of context because it is existing only in this particular chat windows memories, please self anneal where needed to make sure that we don't lose any required context memory. So that'll be what you put in when we're done. And then because we're not going to chew up the entire context window, right? Like, like this little wheel will not be full when we're finished. The one down here that won't be full by the time we're finished, or at least I don't think it will be. I encourage you to try and use it. But then now anytime you open this, you should be able to open up a new chat window and talk to it. It stopped. Scroll up. Okay, so here's the structured plan and initial draft directives. Okay, so now you say, I like the plan. Go execute. And yeah, you're just gonna hit that accept all button a lot because what happens is it's creating a file and then it thinks that file is going to exist. Like it'll create the file, think that it exists, and then it's going to like go back and maybe want to use like not all the time, but sometimes it's going to want to go back and use the file. So you want to make sure you've accepted it so that it it exists. A good example, See this, how it's like red, right? This one right here, that's a file it deleted. So you can see it still because it's kind of like in a temporary memory. But technically that file's been deleted. It doesn't think it's there. It moved it into that temporary folder. So it's just a good thing to have. When this is done, you should be able to like minimize this window. You won't even look at it much and you'll be able to treat this window here like Chat GPT like you were before. And you're just going to get better outputs because you have gone from using Chat GPT to you have now built a software. Right now as this exists, let's say the Analyze Photos MD file and the Generate Scope MD file, let's say those work as we think they're going to. You could create a front end software with a button that says Analyze Photos and a button that says Generate Scope and if it knew where the photos were, and right now they'll exist in the chat window window. But if you had like a thing that dumped them into a folder somewhere, you could create a front end software where a button would trigger the Analyze Photos button and that would go into the executions and run the scripts and do the work. Like you have a software right now you're just interacting with that software. Instead of buttons and mouse clicks, you're interacting with it through chat. All right, so it looks like it's done. So switch to chat mode and then just say if I had some photos and an inspection report to give you, can you explain to me how this works? Like how I would use you? This is going to kind of show you what to do and you can talk to it naturally. Like you don't have to remember exact things. But this is a kind of understanding how it works. You don't even have to follow this process exactly. So what I would just to save this, go to code mode and just say can you save this as a human instructions MD file file? Okay, so it created a howtous MD file. So this has just made it. So you can go and click on the how to use file and you can remind yourself of this later. The last thing I think we do is make sure we like, remove that memory. So when, once this is done, so paste that prompt in, just put it in the chat. So for the time being, I can only use this on this computer though, right? Correct. But you could then host this in a remote environment and then you could give other people the ability to access because, like, instead of opening this folder on your computer, they would open it on a Docker hosted thing, which we'll get there. But I just want to, like, before we start handing this off to people. So I would say that you could test this in this window. Now, like, when I leave, if you've got time, I would encourage you to play with it a bit while everything's still fairly fresh in your mind. And on our next session, I want to see how things went. And the nice thing is we'll be able to review the chats and I'll know, oh, how this thing works a lot better. So I'm going to go. You could use this now to test because you still got plenty of context window. So if you want to add a photo, you would add a photo with the little plus button. I'm mainly just doing this in the ask. Well, yeah, because Ask won't be able to do any work. It'll just be able to make a plan. Okay. And then you would switch to code mode to execute the plan. Yeah, I'll mess with it, but I'll have to find some decent photos if you're worried, like, I would even, like, create another duplicate of this folder on your desktop just so we have it like, like where we left it. But use this folder. You know, just duplicate the folder so you have a backup, but then use this one because this is where you've got all the context and everything. All right, roofers, listen up. Want to build this yourself? Grab the free Step by Step guide. It's your ticket to owning this process. No excuses.