Software Review — Conductor IDE Review

I love coming across new things to test. While Doomscrolling X, I saw Garry Tan post about Conductor. It immediately caught my attention because he specifically mentioned the agentic capability within Conductor. Having tried to run instances of multiple agents myself, I decided to download Conductor and give it a try. Follow along to listen to the build. Also download Conductor to test it yourself. Quick, easy, and most importantly, FREE!

Transcript

SPEAKER_03 0:00

So I came across this post from Gary Tan today, and he was talking about um an awesome agentic um IDE that he was using, and it's not a IDE like Visual Studio Code or cursor, it's specifically made for agents, and it is optimized for codecs and specifically uh Claude Code. So, according to the docs, those are the only two that are accepted at the moment. Okay, so conductor is an IDE. It says run a team of coding agents on your Mac. It's specifically made for Mac, and you can download it, download for Mac right there, and you can do it. And I already got it downloaded, so I'm not going to do that. You go down here and you can read about it. And you can see how the IDE is set up here. And so I knew nothing about this before this morning, and I had to check it out. And it looks pretty simple here on their page. It says, uh, so this is their web page, conductor.build, how it works. You add your repo. Conductor clones it and works entirely on your Mac. Deploy agents, each clawed code you spin up gets an isolated workspace. Conduct, see who's working, what needs attention, and code review. Does conductor use work trees? Yes, each conductor workspace is a new Git work tree. Which coding agent does conductor support? Claude Code and Codex want to see something else, email us. So this is where it's saying, hey, if you if you want to use Hermes or something like that, uh it's not shipped for it right now, but we may add it if you contact us. Uh, how does conductor pay for Claude Code? Conductor uses Claude Code, however, you're already logged in. If you're logged in to Claude Code with an API key, conductor will use that too. If you're logged in with the Claude Pro or Max plan, conductor will use that. So I thought, sweet, I'll give it a shot. So I pressed the download button, downloaded conductor, and it brought up this. Voila. And it was this empty workspace and it was a playground. And so, you know, I said, all right, well, I want to create a test for this. I want to do a test for this conductor. So what's the best thing to test conductor with? What's what's one of the things that we all like to do that we all like to use? We can we can test in our development processes. Uh, it's creating a game. So I went ahead and told conductor because when you create a game, you can, as simple as the game is, it really doesn't even matter what the game is. Um, you can see if it's going to work or not. You know, you you know that if the game it might produce this great game, but you can see a bunch of different glitches in it and whatnot. Well, so that's what I told Conductor here. All right, so check this out. So I came to Conductor and I gave it one prompt. This was a one, this in a simple one prompt. I said, create a fun and interactive game. Use your imagination, spin up as many agents as necessary. You have full creative autonomy, make it happen. So it said, okay, and it went through and it did all of this, and I was watching it, and it's doing all of this, and it did not ask me a single question, didn't ask me for a single approval, and it was going, going, going. Finally, it gets down here. You see where it starts writing the lines of code, uh, 21 lines of code, 104 lines of code, 154 lines of code, and so on. And it gets down here and it says now the big one, the main game file that wires everything together. Write 339 lines, check this out. Write 48 lines, list all game files, start local HTTP P server for testing. So it did a test on on um on its own, said it was good to go. Um, it started the MCP for the Chrome uh dev tool, so for the Chrome browser, and opened it up, and it hit an error and then it restarted and did it again. And then it says the game is running with zero console errors. I noticed it jumped straight into gameplay instead of showing the title screen. That's because the title is only shown in the menu state, but it seems the state transition. Let me check. The issue is that the page load may have triggered an enter key press, but more importantly, the game is working. The ship is centered, stars are twinkling, HUD is showing score, health alerts, wave indicator, and high score. However, I noticed the title screen isn't showing, so it goes on. So it thought the title screen wasn't going when I opened the title. It was working just fine. Um, but it wanted to make sure that it was working. And then it says here the title screen looks fantastic. Cosmic Drift was the name that it named that program. It says uh twinkling star filled in the background, press enter to start. So I want to show you here where this game came from and what it did. Okay, so let me open up a new window here. Oops, not music. All right, we want to open up cosmic drifts, so we're gonna say cosmic and we're going to uh properties save yes, window. Alright, so uh right here. Check this out, guys. So this is my cosmic drift. This is exactly what came up the moment that I first opened it. Remember, one shot prompt into conductor, it did everything. I gave it one simple prompt, not a long prompt, not a detailed prompt. I said, hey, make a fun game, make it interactive, you have full autonomy, make it happen. And here's what it came up with. So we're gonna play this a little bit so I can show you how good it did. All right. Just give me a second here. Alright. Okay, check this out. So press enter to start. I want to press enter. And here we go. Here's our game. And this is a one-shot prompt, right? Oh, oh no. And you can see the the score up there on the top left. And I can even go up and back through the bottom if I want to go take out this asteroid right here. I can continue fighting the asteroids, shooting them, go to the other side. And I thought, man, this is really, really cool. I'm doing, and I kind of got into the game when I started playing. This is like my second time playing it. This is really fun, actually. And so I was totally amazed that conductor was able to do this in one shot. It's amazing. So it even you can see the health score up there, or I have three hearts, you know. I want to see what happens if I run into one. Oh, I lose a heart. You see that? And so I wonder what happens. Oh, I lost another heart. Let's see what happens when I die. If I lose that third. Oh, game over. Look at that. New high score. 6,200. High score. Asteroids destroyed 38. That's crazy. All right, and it's back to it. So now we can try to beat our high score. Oops, lost, lost a heart there. I hit the asteroid. So you see where this is going, right? I mean, this is this is, like I said, a one-shot prompt. And with nothing else, there was no other help from from Claude. There was no other help. I mean, Claude's the one that did it. Claude Code's the one that did it, but I did not go back to Claude Code like I usually have to do. When I build a program, uh you know, usually we'll get the program set up and it will be, you know, kind of the basic setup of the program. It kind of gives you your general architecture and it will run in most cases, but it's very vague and empty. And even when I've done games, you know, even when I've just messed around with games, going and have it creating a game and trying to get it in this state, I have not had that success. So this is by far the coolest one-shot prompt I've ever done, and it was done with conductor. So I took conductor back and I said um I wanted to try something a little bit different to see if what else it could do. So I started working with a Premier Plug pro program, and um it's it is one, let's see here. Okay, so you see here, this was a I wanted to make a Premiere Pro plug-in. Adobe Premiere Pro is a the Adobe's video editing program, so it's a very powerful video program. And so I said, create an Adobe Premiere Pro plug-ins specially or specifically for editing long-form podcasts into clean long-form video. Each long-form video should be should also be clipped into social media clips. The program needs to make all edits, it needs to find the hot parts of the podcast and and make all sorts of different types of contents plus montages. Now you can see here, this prompt right here sucks. Okay. If you know about prompt, this is not a good prompt. This is not uh, this is definitely not ideal, but I was just messing around with um conductor, and I wanted to see what it could do with just a rough, generally very roughed prompt. And so it says, this is an exciting and ambitious project. Before I start building, I need to talk through phase one, discovery with you because there are some critical architectures, and it asked me some questions. Okay, um, what exactly are you using this for? You know, and it says for the uh find hot parts feature, speech to text transcription, engagement scoring, silence filter detection. What's your preference? It asked me, do you want to use AI APIs like open AI Whisper for transcription plus an LLM for content analysis, faster to build, requires APIs per cost use, local models, slower, more complex setup, but no ongoing costs, hybrid local transcription, API for content analysis. So, what I like to do is I like to use a local Whisper. You can download Whisper and you can get a local version. And the the model's not that big. I mean, the biggest Whisper is like five or six gigs, you know. So, and you can get the smaller whispers that are much smaller than that. But for to be able to run the so that's the speech to text um transcription. Whisper is a speech-to-text transcription, works fantastic. I highly recommend it. And you can get it on your local machine, and then you can have an AI come and run through the transcription and do an analysis of that. So that's what I did. That's what I chose to do. So I have my whisper set up, it's the local transcription service that when it's when I finish the podcast and I first run through it, it's going to use the whisper first because it's taking my voice and it's turning it into text. But it's taking everything, it's taking the hums, well, uh, you know how I talk. You know, by now I'm not the smoothest, cleanest talker. And so it the whisper takes everything. And so you're getting basically a raw transcript, which is great. We want the raw transcript, but we also need to get a filtered version of the transcript. And so what we and it and you can get an intelligent analysis of that transcript by having uh a large language model AI come and run through that initial transcription. And once you have that AI, that large language model run through that initial transcription. Now you take it and now you can really make it interactive and do whatever you want because the intelligence from the AI knows what to do and can find the quote unquote hot spots, like I was saying here. So you can definitely use um you can use local AI language models if you want to do that. You can use you know Quen. Um, there's a ton of them. But in this case, uh, I am going to use local whisper and then Claude or Anthropic Claude is the API that I'm plugging into this program, this podcast tools program. So that's what I opted to do. Then it asked me here content types, what does V1 need? And it says, okay, I'm gonna be doing these already. The uh clean edit, highlight detection, social clips, montage builder. But it asked me, well, what about multicam support? What about captions and subtitles, aspect ratios, b-roll markers to suggest where to insert B-roll clips? And so what I told it to do, I said, I want you to do all of them. And then it asked about podcast format. Am I using video, or is it just audio, how many speakers, typical episode length, and output target? So here was my response to it. I said, I'd like to use local whisper model for transcriptions. I will provide anthropic API key for analysis. We also need to add multicam support, caption subtitles, aspect ratio presets, and b roll markers. I do podcasts with video, typically solo or two speakers. Episode length is usually 30 minutes to 90 minutes. Outputs are all of the above, referencing the outputs as far as um the social targets right here YouTube, TikTok, Instagram, X, LinkedIn. And so it said, all right, here it goes. And it and it created this context here. It says editing long-form video podcasts is tedious and repetitive, and it gave an architecture um overview, and then the tech stack. And so then it started working, and it worked, and it worked, and it worked, and it worked, and it got down here and it said it was finished. And so this is where I'm at right now. I went and uh but said CD into Memphis. When you create a conductor uh session, it is going to or workspace, it is going to give you a it is gonna create the workspace with the name of the city, I guess, because the first one that I created was Dubai, and then the second one is Memphis. So um it's saying CDN to Memphis, then run the the bash install script, which I did, then uh run the backend server, which I did. But when I ran that, the install script properly asked me, okay, you're run you want to run Whisper, which which model do you want to download? And so I chose the biggest, and which was only like five gigs. And then it said, hey, you need to enter your anthropic API key. So I also'd enter my API key. And this is in the shell terminal that I had opened on my computer. And when I entered the anthropic API key, it rejected it. It didn't, it said it was an error, it didn't reject it, it just said that there was an error. And so what I did is I came back and I told it, I said, two issues arise. The anthropic API key was not accepted by the program. I copy pasted anthropic API key, but it's not working. Also, Memphis is not an extension in Premiere Co Premiere Pro. Can't find it error invalid configure. Oh, and then I said uh I copied and pasted the error that I received in terminal, which was this error right here. And then it says, okay, each subfolder needs a manifest, and it goes down and it recognized. I don't see where that text was. Maybe it's up here. Oh, right here. Yeah, it says two concrete issues. Let me fix both. Issue one, the backend isn't reading your.env file. The uh TSX runner doesn't auto load.env files. We need to add dot and uh env support, and then the issue two, the uxp extension can't um just be simlinked. Premiere pro needs to load uh needs it loaded via the UXP developer tools, or we need to provide a CEP fallback since UXP plugins loading is more involved. So it said, let me fix both, and it started doing that. And so this is where I'm at, and you are going to walk through this with me and see if we can get this program running. Because, all right, here it says to CD into Memphis and run the back end again, but I need to make sure that the all right, the C CEP extension is complete. Here's what was created, fixed, the extension CEP, the index.html, the full panel UI with all six tabs, the montage, da da da, Adobe's CEP communication bridge library, fixed, remove duplicate, extension bundle versions, attribute that would cause XML parse failure, uh, fix the script, the install script. Alright, so we're gonna give this a shot. I am already let me clear this real quick and let me pull up this terminal so you can see what I'm doing here. Alright. Okay. This is iTerm two. ITERM2. I love iTunes. Okay. Alright, so we see here we have this terminal. Oh my goodness. Oh my goodness. Alright. Are you with me still? You see this terminal right here? This terminal window is the one that we're going to be working in. And so I am actually going to bring that right here so I can see this. And we are going to where are we at? Alright, we need to run the back end, so we're gonna do npm. And you can see here I'm already in the the um the section, the right directory. Uh npm run backend. Let's see what happens. Alright. It's saying it's fired up on 9876. So let's see what we can get if we go to there. This will be my first time trying it. We'll give it a shot. And I got a 404 status that you can see in the terminal. And you can also see if I bring this up. Oh no, not right there. Um well, I got a uh in in my in my web browser, I got uh a message not found. There's an error. We I got a 404 error, which is not good, so we're gonna go back and read through this terminal and see what's happening. We had a git, we had a request completed. I wonder what is happening and why it did not do that correctly. All right. So what I'm going to do is I want to come back here, and that from right here. So I'm just going to copy it, command V, paste it, install it. You see what's happening. This is a oh, by the way, this is absolutely free. This is here's another thing about this conductor. This it's free. So other than you know, using your Claude Code or Codex, your the program itself, the conductor program that's allowing you all this agentic capability is um or multiplying the agentic capability is it's free. So install local whisper. I'm going to say yes here, and I guess it didn't install last time. So I'm going to go with uh the large ones. So five, you see here it's saying install local whisper. Uh would you like the tiny one at 75 megabytes? The base, small, medium, or large, which is 2.9 gigs. I said five something really 2.9 gigs. Um, so number five. All right. So you see here. I want to read through this because I haven't I haven't ran this program before. So whisper.cpp already clone, pulling latest, already up to date, building whisper.c. Detected Apple Silicone, which yes, that's what I'm running on. Uh CMake. I had to install CMake to get all this stuff running, but the program, the the install script that it that it um ran had cmake with it. But when I ran the install script the first time, I was reading through and it didn't download CMake for whatever reason or didn't get it installed. So I went ahead and did it manually. And so let's see here. Update the version, cmake warnings. So these are just our warnings. What's this here? Checking for ARM features using flags. So target whisperer setup complete, whisper binaries there, the model file is there. Add your dot env here. Um add to your dot env here. Okay. So I'm wondering if I need to run this for the dot env for the path to be there. We'll see, or should have done the uh back end. That answers my question. Installing Premiere Pro CEP extensions, enabling unsigned CEP extensions, uh extension installed at, and it tells me where it installed. Installation complete, quick start, start the back end, cd Memphis and npm run back end, open or restart premiere pro. So I have Premiere Pro over here. I am uh it's open. I'm gonna close it. Okay, I closed it. You can't see, but just trust my word is closed. Um and window, oh, open or restart premiere pro. So let me open it again. And let me bring up a new project and yeah, let's see here. Let's see if this. Oh, look at this. Okay, guys. Okay, guys. I gotta show you something here. It at least worked. And I'll show you. I will show you my whole screen, and it is my bin Q monitor that it's on. So okay. Y'all see this right here? So here is my here is my Adobe Premiere Pro. And let me fit it to the screen for us. There we go. Okay, so this is Adobe Premiere Pro right here. And what I'm trying to do with our Ben Q, or sorry, our Ben Q, what I'm trying to do is build that program, the podcast program that we've been messing with, then um conductor, and look at what it so it says to go to window extensions, and look right here. You see this? This is the extension. I want to click on it for the very first time. I don't know if it's gonna open. Oh, it opened over here on this page, on my other monitor. Let me drag it over, and it says disconnected. Okay, but look at this. At least we have the window pane up, and it's showing us analyze, transcript, clips, montage, export settings. Of course, the other thing that it's saying right here is that the back end is not running, so I am going to show you what I'm doing. Actually, I need to just give me one second. I have to get out of this full screen so I can bring my terminal over. All right, so I'm gonna control C to close this out. I'm gonna clear this, and I'm actually going to go to my home directory and clear it. So now I'm at my home directory, and I am going to follow this right where we need to be. And if I can remember, I think it's conductor, uh, workspaces where it is prod or pro sorry, pro pod pro one. That's a tongue twister. Pod pro one, I already did it right there. Okay. Enter. You see Memphis, Memphis. Okay, so now here is where we need to run our. Do I need to cd into back end first? It told me up here. I probably do. No, it just says run back end. Okay, so CD or sorry, uh npm run backend. And here we go. So now what I'm going to do is I am actually going to restart Premiere Pro to see if we can't get it to recognize the plugin, correctly. All right, there we go. We're just going to do a new project, say yes, replace it. Window, extensions, Memphis. All right, back and not connected, start Memphis server. Well, that's what we're doing. Hmm. So I'm I'm just going to go back here. We're walking through this together. And we're gonna go back and we're gonna tell the agent that was able to get the oops, oops, oops, oops, oops, oops, oops. Was able to get the extension installed and premiere. However, when the extension opened, an air backend not connected, start the Memphis server was shown. However, I have started the server and it still does not recognize the back end server. Please investigate and fix. Let's see what she does here. All right. So I'm gonna give it a second. It could be that um let's see. Let me check the configs. So what I'm going to do saying the dependencies are installed and the.env exists. Let me try starting the back end to see what happens. Your.env is your environment variables files. That's where your um your secrets go. So if you're using a program and you have passwords and you had API keys or something like that, you don't want that in your code. So your.env file is where you put your your secrets at. And then you import that into your code. And then when your code is running through, it references the.env file that's hidden from the main uh code, but it can get all the passwords and API keys or any secrets that you need stored. So it says, now I see two problems clearly. Let me fix them. It says, I see the issues. The manifest needs additional CEF flags to allow private network requests. All the panel uh JS needs to handle older Chromium APIs. Let me also add a.ce config approach and switch from 127.001 to localhost since some CEP versions treat them differently. And so it's working. We love to see it working. Now let me add a robust fallback. So I'm gonna go ahead and I'm gonna go ahead and close Premiere Probe because I'm not gonna have to to restart it. And let's see, I'm also going to stop this server by command C and I'm going to clear it because I like a fresh little terminal shell. Yeah, so what you see me here uh doing here is we're getting this program running, we're using conductor, and uh conductor being the great, great agentic IDE. I guess that's what we'll call it. Uh I don't know. I I just literally came across this this morning and um conductor that is, and uh since using it, been using it now about an hour and a half, I guess, two hours, uh messing with that game at first. That of course that game was a one-shot prompt and it built it perfectly. And now I'm building a much more robust tool, so it's working through some some issues that it's had, but these issues are just what you have to do, or you just have to work through if you want to really build anything. So um, for those that are doing this, uh, you know that this is just part of it. But if you're new to this, let me help by saying you're going to run into errors all the time, and you have to have the perseverance to work through those errors. And at first, when you're getting errors and you might not even know what they're for, right? Like it may tell you that you have an error over here, and you're like, Well, what's even over there? I've never been over there. How do I what am I supposed to do? Well, what you can do is you can take that error and copy and paste it and go feed it into Grok or go feed it into Chat GPT or whatever your favorite LLM is, and you can give it context about what you're doing, and then the error that you came across, and it will tell you, most likely, instructions on how to fix the errors. So it's a really good way to be able to bounce back and forth. In this case, I'm just using um conductor, the agentic IDE internal development environment, and I'm not even having to use uh grok up to go and check errors or anything like that. I haven't had to use it yet for this, and thankfully, it looks like it is already fixed the the errors that we were having. Uh it says here's a summary of the changes. Okay. Okay, to test, restart the back end, npm run backend, and then restart premiere pro. So I'm gonna do the back end right here and run. Actually, let me make sure that it is right there. Yep, you can see it. So you can list what is in the directory, right? And and this one is the back end that we're of course wanting to run. So I'm gonna run uh npm run backend, and it says, looky here, look here. Give us some more space. Okay, npm run back end or npm run dev with the flag workspace back end. Um we got a deprecation warning that is normal, I believe, for the puny code, although I do not like seeing that. Um, so what we're gonna do is we're just gonna try to run Premiere Pro. Let's do it. I have it open here. We see that we got it open. Gonna do a new project, just a fake untitled project. Yes. All right. And Windows Extension Memphis. Looky there. Look at that. Y'all see that? Wow, this is amazing. So um, the issue that we were having where it was showing that the back end was not connected, the server backend wasn't able to connect to um RPMere Pro extension, or really it's vice versa. RPMere Pro extension was not able to connect to the back end. So all I did is you see that I went back and I told Conductor right up here when I came to that air, I said uh that was the first air. So this, so so this I've had two airs, two little speed bumps, and they've been very simple. The first one was with the anthropic API key, and then um um the Memphis extension, the Premiere Pro extension itself wasn't visible for me. So those were the only two errors. The the anthropic API key was fixed initially after the second run, but the extension issue uh took a little bit for it to fix, uh, but really just just twice, right? I told it the first time, hey, it's not working. The extension said, Oh, hey, I fixed it. And then I come back to it and says, Oh, actually, it was able to get the extension installed in Premiere. However, when the extension opened, an error back end not connected, start the Memphis server was shown. However, I've started the server and it's still not recognizing the back end server. Please investigate and fix. So it goes through, and now I have this. So if I'm rolling here, if we say that we want to do a solo speaker, y'all want to run a test here? Let's do a test. Let me go and I'm gonna just bring in a video. And I don't even know what this video is. Oh, that was me doing just some guitar playing, so I don't I'm not even talking there. Um let me do final video. Oh, final video. Sorry guys. This is uh the system that I'm working on here is my old computer, and as you can tell, I am not um organized whatsoever. Let's see. But on my new computer, I learned a lot from this one. So on my new computer, I um I definitely okay, let's see here. Zb pod final video. No, clips? Um let me check if I have something in trash. Okay, what is this here? Let's see. I don't even know what this is to be honest. But it's me talking and saying something. I'm gonna open this. And see where it takes us. Okay, so I have this little 18-minute video here that I am going to import. And that's on my desktop, so I'll show you here. Oops, desktop. Uh yeah. Okay, so we're gonna import this video, and I'm gonna bring it on this timeline so we have a sequence. Oh, I know what this was. This was one that didn't work. So this was one that froze. So this is not going to work. This is a bad example. I'm just gonna remove this. Let me try to find another one. Um put this stuff in trash, and it's like, hey, it can't be in trash if you wanna play it, so it needs to be here. And that's not either.

SPEAKER_01 41:51

Okay, I think this one was one that we can do actually.

SPEAKER_03 41:57

This was a podcast that I did with my. Oh, it's myself. Okay. Let's just see here. We're gonna drop it. Alright, let's drag this over here. And let me make sure my OBS is gonna bring in the audio. Okay. So I think you can hear. Let me pop this. So I hear what you're hearing. Okay, so you see here this is a uh this is a podcast that I did recently. And um you have to have good ideas to put into action. But if those thoughts are just I want to turn that up a little bit. So this is I was recording into my road, but you can see that the audio is a little low, but but that's okay for this purpose. What we're gonna do is we're gonna try it out and use our extension for the first time. And let's just see how it does. So we're wanting to run, we have one speaker and video file path, um, path to podcast. We really should be. I'm not understanding. So it's almost saying like it wants to see the path to my my folder where the video lives instead of just processing. I'm just gonna try to do it right here and see what happens. Uh, enter video file path first. Okay, well, I actually want to do it on the sequence, so I'm going to just follow this instruction here. I'm gonna erase that, and I'm actually going to erase that. And what we will do is we'll find this path right here. So uh copy as a path name. There. Okay. Solo speaker. And it defaulted back to two speakers. Uh conductor will look this. Okay. So I'm just gonna see what it does. Uh no JSON fountain and whisper output, raw output. Okay. Well, first of all, you're okay. So we have some issues still. We've we've we've gotten better, um, but we still have some issues. So we're gonna go back to conductor and we're going to say um a few issues. First, the plug-in extension plugin requires a video path to analyze. I want the the extension to analyze the video in Premiere Pro without having to import the path, the video file path. Instead, the program should consider all videos, video audio on the sequence timeline and edit accordingly. There may be one then, or sorry, there may be more than one video on the sequence timeline, and the program must account for all content. Uh, what was the other thing? When oh it said something about the JSON wasn't recognized or something like that. Uh no JSON found in Whisper output. Okay, so we need to tell that. So one thing I like to do kind of when I um when I'm when I'm talking to these AI, these agents like this, I like to separate my thoughts a little bit. If I kind of have, you know, kind of like going to a new paragraph type thing. I just like to separate it with a few um forward slashes, and and that'll just kind of break it up. So there was another error, another two other errors. Geez, I can't type when I tried to edit using the extension. First, the program the extension. First, I set extension to one solo speaker, solo speaker, and it immediately defaulted to two speakers when I clicked. Is that the analyze button, or what is that button that it has there? Analyze podcast. Um click analyze podcast button. Secondly, oh I'm gonna tell it make sure it uses it uses the required the selected selected um speaker setting. Speaker setting. Make solo speaker the default. The default. Secondly, it returned an error saying invalid JSON format for whisper. Uh please investigate investigate these issues. Write test scripts as necessary. All right, let's see what it does here. So we're gonna tell it that uh hey, we had these issues, and yes, we we got it up and running, but um, but there were still some issues. So this is the thing. You take uh three steps forward and then you take two steps back, and then you take another three steps forward and two steps back. Sometimes you take three steps forward and three or four steps back, and you just try to keep taking steps forward, and eventually you'll get to where you need to be um where you're happy at. Let's see it saying check whisper CLI help for output, the key finding, the whisper CLI, the newer whisperer.c uses output.json. Oh, I missed it. Where did it go? Okay, I love to read this as it taught you. You can learn so much from just reading what it's saying. Check if the F flag exists. Okay, it does exist. Now I have a clearer picture of all three issues. Let me fix them all. Issue one. The whisper.json output, whisper Cli writes JSON to a file, not uh std out need to use or standard output, not uh need to use output json full plus output file, read the result in JSON. So issue two speaker count. The HTML has selected hard coded on two speakers, and the reading logic may not capture the select value properly. Issue three video path. It's firing up a server again. You see it over there right here. It's doing all that on its own in the terminal, so um, all this is just rolling. Incoming request, it shows it here. It's testing. We got a 200, which is good. Uh, another 200 code, good. Okay. Okay, now I need to update the state to track media paths and speaker count and replace the analyze tab rendering. Now replace analyze tab rendering to show timeline media instead of a manual path input. And it's still rolling, so we're just gonna watch it. This is kind of the most exciting part, like this. This is like the thrill, right? It's the suspense builds, and you don't know what the output's gonna be. You think it's gonna work, you hope it's gonna work, and usually in my experience, it um I guess most people's experiences when you first are getting started and the program, whatever program you're building, it's really never going to work properly the first, second, third, fourth, fifth, sixth, seventh, eight, ninth, tenth time. It's the 11th, the 12th, the 13th, 14th time that you do it, and you're like, holy crap, it works. You know, you kind of get you expect that it's gonna work, you know, the first few times, and then you just hit the air, you hit the air, and you go back to it. You try to fix the air, and it doesn't fix it, and you go back to it, and then you go back to it the 15th time and you click it thinking it's not gonna work again, and bam, it works, and it just kind of like hits you in the chest, you know. Okay, we gotta you see this right here. It's saying we are it's trying to listen, but it's saying that it's already in use. It's a very interesting address. But we're just gonna let it roll. We're gonna let it roll and do its thing and and hope that it's doing the right thing. Hope that it's uh trust that it's doing the right thing. See there, it's messing with the transcription. Um file, transcription. All 11 of our tests passed. The only failure is an unrelated test file inside the whisper directory. Well, that we'll see. It's saying there's an error with the whisper, but it's an unrelated error with the whisper test. So no vite test config exists. It uses default. The whisper directory has a test file that vite test picks up. Let me add vite test config to the root to exclude it. All 11 test paths cleanly. Okay, so does that consider the or does that also include the whisper file? I'm not sure. Let me fix the model path resolution. Okay, also the model file. Okay, here we go. Yeah, it looks like we're still getting some errors, but there's a bug. Yes, there is a bug. Is now async, but the call site to run whisperer cpp doesn't await. Let me fix that. Oh, look at that. I think it worked that time. Yeah, we got 200, so hopefully it. Okay, oh, there's that error again. Also, now verify the backend compiles and tests still pass. There are pre-existing type errors not caused by my changes. Well, you built this, dude. Let me verify my changes. Don't introduce new issues by checking just the transcription file. All 11 tests pass. Let me also quickly fix those two pre-existing type errors so the back end compiles cleanly, which is of course what we want. And it's still not doing that. So now fix the unused ZM port. You see, we still have this back end error taking place, and still taking place. It's truly unused, remove it. So it's saying that it's going to, you see this? Looks like it's saying it's just going to remove it entirely, but whatever that test file is, or I'm assuming.

SPEAKER_01 56:25

Okay.

SPEAKER_03 56:27

Test these fixes. Well, okay. Restart the back end. So I'm going to control C here. I'm going to close Premiere Pro. Um I'm not going to save this. Alrighty. So restart back end. I'm going to do npm going back end. And once we have that running. Concerned about this. But we're just going to roll with it. See what it does. Fire up Premiere Pro. We're going to start a new project. Let's do this new project. And I'm actually going to close. I guess I'll leave it open, but um need to import the media. And I think it's right. Chia. We're going to do this at. And we're going to bring it over here and make sure that this is it.

unknown 57:34

Many, many people are jumping on and seeing the power of OK Claw. And um, so there's there's been some people that that uh friends and family that that I've hooked up.

SPEAKER_03 57:45

Okay, so um so here is our timeline. Here we we imported our video. It's about a 30, almost a 40-minute long video, I guess. Um and notice here, I actually didn't notice if this was there before. This could have been there. I don't I don't think it was in our previous version, but but I asked it, I said, because it was only wanting me to enter a video path, and maybe it had this, but I was thinking to myself, well, if we're doing a podcast today, we're gonna have multiple, we could have multiple videos in here, not just one path, one video file stored somewhere. So um, so what it did is it implemented this detect timeline media. Let's just see what it does. Okay, it detected that sequence right there, the media files on timeline. Okay, so low speaker. I'm gonna do analyze podcasts, extract audio from video. Okay, we got another error. Whisper exited with code three, whisper in it from file with params, uh fell to initialize whisper context. So, what I want to do here is I'm just gonna copy this and I'm going to open conductor again, and I'm gonna go back to conductor and tell the agent that tried testing again and pre mirror pro ran analyze podcast and received this error. I'm just gonna paste it right there. Wait, it didn't paste, did it? Uh let me let's see, copy and let's see if this will paste. There we go. Okay, so let's just see what it does. And it's gonna think for a little bit, and then it's gonna come back and say, Hey, I found something. The model file doesn't exist at the path. Earlier we saw the actual model is okay, so it's telling me that it can't find the model. And by model, we're talking about the whisper model, the transcription service. Uh, but the config says whisper base model is trying to connect to whisper ggml, uh, which doesn't exist. Oh, wait, you know, and so this brings up a question for me as far as where do I have my stuff stored? Because um, let's see here. I'm over here looking at my own stuff and trying to figure out where I have I stored local AI models. Uh would that be in llama? Would I have stored that in llama maybe? There's the G stack G stack Gary Gary tan. Okay. Local. Uh no, not Ben. Uh that's not it. Um there a llama models blobs. Uh no, that's not it.

SPEAKER_01 1:01:36

Look at that. Local AI models.

SPEAKER_03 1:01:41

But I don't have any in there. Okay, what happened? What happened to everything? Um, we're just gonna let it roll and hopefully it will fix everything for us because we're just having an issue with the whisper transcription. It's saying we have a corrupted file. So it's gonna say it's gonna download a real model and fix the path resolution. First, let me download a proper base model using whisper.cpps, download script, and we're rolling with it, and we're rolling with it. Uh verify download model size. Oh, the model downloaded already. 141 megabytes. That's a real model. Let me fix the well, I want the big model though. I don't want the small model, but we're just gonna do this to test it anyways.

SPEAKER_01 1:02:52

And but look, we're still getting this air, so I wanna see if there is an audit.

SPEAKER_03 1:03:15

What type of eleven tests pass. Now let me let me also do a quick sanity check. That the model resolution actually works with the current directory structure. Ah, real bandurr is dot whisper and it goes, but I used a relative path. Let me test with the absolute path. Verify model, the models candidate from the real binary directory uh resolves correctly to the model, 141 megabyte model. The resolution logic will find it. Now let me rebuild the back end. So I still haven't seen the backend rebuild. Um restart the backend server and try again. Hey, we're gonna try it. So, all right. Let's do it again. npm run dev. We still got that error, so I'm not too confident that this error will be resolved. Let's see, we're not gonna save the project. I could have saved the project, I probably should have saved the project, it will save me time. But I always just like starting fresh because you know you're getting a fresh start. Alrighty, here we go. Bring this to the timeline here, and we're gonna detect timeline media, analyze podcasts, and I suspect we'll get a whisper error right here. Any moment now, 18%. Although it looks like it's going a little bit further than it did before. Let's see if it's actually transcribing or if it's just hung up on something. You never know unless you have detailed logging, which in this case we don't have detailed logging, so we're gonna hope that it's transcribing whatever I said, which there's really no telling what I said.

unknown 1:05:54

It's the same way that you kind of do with other chatbots, right?

SPEAKER_03 1:05:58

Like the other you see the So we got another, we got another model or another error. So this error is that the Claude Sonet45 is not found. So we worked through. So this is good news. This is good news. The good news is that we worked through that whisper error, and we are now facing a different error. But hey, remember what we talked about. We take three steps forward, two steps back, and we're okay with that. We're even okay taking five, ten steps back. That's okay. Sometimes it just all it all messes up, right? And just pick up the pieces and get started again, or just throw the pieces away and start over. That's okay. I'm gonna copy this. I want to go back to conductor and I'm gonna say looks uh ran test again. It appears we worked past the whisperer error. However, a new error arises related to anthropic API. Please fix. And then I'm going to say the error is like this. And send it. Send it. Let's see, man. I need some like I need some um, you know what I need. Oh no. That's not what I need, right? I mean, come on, man. We need like some. What is this? So when I'm talking, I can say we got a big plague.

SPEAKER_00 1:07:45

You like that? Alright, or maybe if I just want to sound like a little silly schoolgirl, I can talk like this. Alright, I'm having too much fun over here.

SPEAKER_03 1:07:57

Or if you really need to hear me, I can yell at you through a megaphone. Oh, that's what I need to do when I when I um when I come back and I'm talking to my agents, I say, correct the issue. Fix the issue. All right, what happened here? Uh please connect your discs. Oh, whatever's going on. All right. The model ID, Claude Sonet45, was a non-existent model, likely a hallucination. Oh, hallucinating, huh? From when the code was first generated, the correct Claude Sonet is Claude Sonet 4. No, I do not. What is this saying for the anthropic models? So what model would anthropic model IDs follow the pattern Claude tier version date. Claude tier version date. Claude tier version date. Well fixed. The model ID was that Claude Sonet restart the back end and try again. So it's saying that it's removing the five from the code. You see that it's saying Claude Sonet45 here, and here it's saying ClaudeSonet45. What I am going to do, actually, is I am going to Google um Claude Model naming convention docs. Okay. Let's see. So what I want to do is even though it has the capability to go through all this stuff, I want to say we're actually in March. It's actually March 30th, 2026. And those models you're referencing are really old models. Please see the Claude docs for reference. Um I want to use Sonnet 4.6. So I'm going to give it this, and hopefully it goes and it runs through these docs real quick and it can find whatever issue it was having and correct it. Because it's saying that um, you know, the clawed models use a date um and maybe they do technically um under the hood, but I have never seen a date used with a clawed model. So that's kind of new to me if that's the case. Um you see here uh fixed model IDs. Model ID updated to ClaudeSynat 4.6, the current ClaudeSonate 4.6 API identifier per the docs, restart the back end and try again. Okay. So we're getting all 200 here, which is good. And what up it is doing, so I'm going to restart the back end and I'm going to clear it, and then I want to try it again. Let's run the back end. Let's quit Premiere Pro. I want to save the project this time. Hopefully, it'll save us some time. We're going to open and we should hopefully see our untitled project right there. We'll open it. Alright, here we go. So we're going to have it detect timeline media and then analyze the podcast. And we made it up to 18% right here last time, and it was transcribing, and it transcribed for a little bit until we got that model error saying the Claude Sunet model was not recognized. And I think hopefully it was because it was just um a faulty, faulty model description by adding the year, month, date to the actual model naming nomenclature where Claude, um, you know, or Anthropic, they name their models Claude. Then the model number. Oh, look at that. We made it past okay, okay, this is good. We're at 52% now. So hey, we made it three steps forward again. All right, now let's just see if we can stay it three steps forward. Then if that that's the case, we'll be like 30 steps forward, right? All right, it's processing, it's going good. It's going good.

SPEAKER_02 1:13:17

Okay, well, you you came back with this.

SPEAKER_03 1:13:20

All right, so we're gonna let it process. Now it's analyzing with Claude AI, so it's going through and it's saying that it did the uh apparently it already did the whisper transcription. Now, reminder, the whisper transcription is being done with a very, very small whisper model, so it's probably not good or it's not gonna be the best quality transcription, but that's okay. We're just doing this for testing purposes to make sure that we can get everything in place. We don't need the whole, you know, if we're if we're just testing this, and you know, you could download a big whisper model, you could download these big AI models to do all this kind of stuff, but you're really trying to do it kind of with the bare minimum, right? Instead of running a three gigabyte whisper model, you could run a 145-megabyte whisper model. Well, that's much more efficient, you know, not just it's on your it's on your machine, it's taking up less RAM, so it's just a cleaner run, faster. Um, but it's also going to have a little bit less quality, as you can imagine. So if it you know doesn't come out with the perfect transcription, we're okay with that. We're just wanting to make sure that that the framework of everything is working properly and uh that the that the program is working properly. And then when we come back, we when we get it fixed, when we get exactly how we want, then we can come back and upgrade with a with a bigger whisper model. Or if I was developing right now using um maybe a local AI uh LLM like Quen, um I would use Quen to do that. Uh Quinn would take the place of anthropic Claude, but right now I'm just using I'm using the API from Claude. And if we look over here, you see that we have our very first successful run here. And Memphis is saying that's the name of our plugin. Um analysis complete, make more money with AI agents. So, this is what I was talking about then um in this whole video, right here. It's just, you know, I was talking about AI, you know, agents, agenc programming, open claw, clawed code, codecs, that type of stuff. Um, it says, make more money with AI agents, how to get started with open claw today. Uh, Zayn Bitton, that's me, uh, breaks down how AI agents, especially OpenClaw, can act like a team of tireless employees working for pennies on the dollar. From his brother's real estate dashboard story to Chrome's to Chrome 146 game-changing browser update. This episode is practical, energetic guide to taking action in the AI revolution right now. Hey, that's pretty good, right? I mean, like this is uh that was actually a really, really good description of of what the video was, you know, kind of just summed it all up into one little two sentences, three sentences, whatever that is, right there. Very simple, but it did it without, you know, sounding, you know, like crazy AI and that, you know, all this hyperbole stuff. That was a really good description. I'm very happy with that. Then it says here, but what I need to do, you know, this font, of course. I I'm gonna go back and change this, make this white. Um, but it's saying highlights, there's 10 highlights, 60 clips, cut 16. So I don't know if that's saying it's gonna cut 16 of the thing out. If we look right here, I'm at this is about a 39 minute and right there, 39 minutes and 29 seconds, uh, and 27 frames. Um that's how long this video is. So we're we're at it 39 minutes and 30 seconds. So it looks like it made these tags. One tag AI agents, open claw, making money with AI, taking action, large language models, API rate limits, open source AI models, agents of workflows, productivity. Now look at this. Apply all edits to timeline. What if I go up here? Oh, this is the this is the transcript. So you can see here when I'm first starting off the video, so this is a whole video, right? And like I turn on the camera, I'm walking in, I sit down. So this is what's happening in the first few seconds. And if I zoom in here, we're at 13, 14, about to be at 16 seconds, and it looks like I don't really start talking until thirty seconds or so when I say test. Let's see.

SPEAKER_01 1:18:17

So we see there here. Okay, so I'm gonna go up here.

SPEAKER_03 1:18:27

Okay. And it gives us time stamps. You can see. Oh look at what look at wait a second. Are these words? Look at this. These words are linked. I can't whatever it is. I have a business and I, you know, wow, look at that. And Google will give me a results. You see that? Look at this. I didn't even ask it to do this. And Claude using uh conductor, conductor, the agentic AI, uh, the agentic IDE with Claude made this program and it did it in such an awesome way that in the transcript, it could have just done a dummy transcript with you know timestamps, and I thought that that's hey, I got the transcript here. I was really happy with that, but it made every single word linkable. So now we have internal links inside the program to where each word is linked to the timeline. So if you see right here in Adobe Premiere Pro, I'm at seven minutes and 35 seconds. That's why I'm at on the timeline. But if you look right here, it's kind of hard to see, but it says 26 minutes and one second, where I say in the transcript, it says we don't know uh what we don't know. Something. Oh, so our agents don't know what they don't know either. But the good part is, and I go on to say whatever. So I'm gonna click on we, and when I click on we, it's going to go to this market around 26 minutes and one second, and we're gonna see that change right here. Let's see if it works. Look at this. All right, where's that? Uh we right here. All right, one, two, three. Look at that. Wow. And so then we don't know what we don't know. So our agents don't know what they don't know either, but the good part is is they know a heck of a lot more than we know about. Look at this. This is amazing. Okay, so that was that was the transcript. And and this, guys, remember what I was saying? This is a this is a small whisper model doing this. So, I mean, you know, I was talking about how you're going to get better quality, better results for the larger whisper model, you know, the 2.9 gig whisper model, which you will get better results. But we're seeing here that the results that we're getting for this real small whisper model, I think it's 145 uh megabytes or whatever it is. This is fantastic, folks. This is fantastic. This is totally workable. I mean, I am very, very impressed with this. Okay, so clips, highlights. Uh, we have a clip here. Action over thought, the baseball analogy. I don't even remember what I was saying about baseball. Uh, AI agents are the thing, not just the next big thing. Uh, AI agents are like five to ten employees working for you, I guess, is what it may say. Oh, yeah, I can stretch it out for pennies or working for pennies. Uh, find your pain points and build solutions. Wow. Okay, so I just want to click on one of these and just see what happens. Find your pain points and build solutions. All right, so I actually jumped over to Claude Code and I worked through this um issue that I was having, and Claude Code fixed it up for me. And you can see here that now we have our clips, they save perfectly, and I was able to export some, and um, it was really cool. So I I I got this program, you know, just giving it this um prompt to do a little podcast or a plug-in for um, I'm sorry, it's technically not a plug-in, right? It's a extension for Adobe Premiere Pro. And in this case, our extension is called Memphis because when we were working in Conductor, the original program that we were in, oop, that's telescale, they kind of look like in conductor here, um, our agentic IDE. Uh it named it after the city of Memphis. So I'm fine with that. And I'm, you know, just testing this program just to see what it can do. But then this in a couple hours we're able to get up and running a um a podcast um extension for Adobe Premiere Pro. So that's really cool. And it's helpful because it's gonna save a lot of time. And, you know, whether it might not be the perfect edits, you know, you'll still want to make your edits, of course, but this is going to give you a really good starting point. I mean, if you got an hour, hour and a half of content, I mean, heck, even 30 minutes, then um you can break up 30 minutes, or this program can by using the whisper for the transcription and then using the Claude for the reasoning behind it, then um then you're going to be able to come up with a with a program that actually can do something. So in this case, we have our different videos here, like this and here.

SPEAKER_02 1:23:52

You've probably heard about it, you've probably seen the little lobster with the claws, but you have no clue what anybody's talking about.

SPEAKER_03 1:23:58

Okay, or maybe I can go to this next clip, and whatever I'm talking about. Active work history, work cases going on, active cases going on, and what it does is of course, every case is different. Clients A through Z, right? Everybody has a different case. Everybody's in different parts of the case, and no case is any is okay. I can go to the next clip. Oh, that's the same one. Yeah, the set knowledge is another clip source, which is I mean, this is so awesome. Um, this this is so cool because now what so what this will allow me to do is we have um this program was able to break this podcast up or break up this long form video, it can be any type of long form video, right? Like, and it will do it intelligently and it will do it in a way that uh you'll have a useful product to to put out on your social media. So now what I'll be able to do is I'll be able to add um some different edits into it. I want to get the um it was doing the captions. You can see here it was layering these captions, but really what you can do in Adobe right here, you can actually just go to well, right up here, uh create captions from transcript because Adobe actually does a transcript already, and um I think it's gotten better over the years. You know, a couple years ago, I think there were more errors in it, but it's gotten better, and so you can go to captions, then create captions from transcript, and what I like to do um bring the characters. So if you want to do like one, if you want to leave it um, if you want to have like sentences, you know, this is your length of characters, the maximum length, and so I bring that down to like 12, and then the duration for how long it stays on. Um I think that's what it is. I think it makes it go fast, but I bring this down too, and I don't mess with that, and then I like the single lines if it does even break it up into two, which sometimes it does, then create captions. And you'll see right here. So this is my original clip. This is a 39-minute one, and you'll see right here that um there should be some captions pop up or some orange things. There's my selected tool. There we go. You see that? So I'll zoom in here. Oops. So you grab the little bottom thing and you can zoom in, and it will show you the different tracks for the captions. So right now we're at this point on the the timeline, and that's why you see it says that that you'd I just said that you've and if I roll over here, so then what you can do is um you can want to close this out here. You can you see how it uh brings up the the properties of the in this case where in the we have uh this text selected for bend times or or the the captions, but it says bend times. You go over here, you can make it smaller or uh bigger or smaller by the font. You can if you want it bigger, I actually want it bigger than that, you know, like that, and then you can change the font if you want different fonts. I like just a simple font, easy to read. You can change the color. If you want to do that, you can change the color. Um if you have a specific color, you can um get the hex code and put it in right there. Or you can also use the color picker. And you can go and select something. So if I wanted to match it to maybe this lava lamp over here that I have going, I can do that. And you see how the lava lamp has different colors in it. I'm gonna go with just kind of that real that color. Just a pink color, right? Why not? Um okay, so we'll just say okay. And if we want to put a stroke around it, that means around the letters themselves. And of course, you can also change the stroke color if you want to. Um but you know, you can get as creative as you want, and you can make all these changes over here. You can put a background. This is for the background, this is the opacity. Um, this is how big it actually spreads out. And then this is the corners, the rounded corners. Let me see there. I want a little bit more opacity. And of course, you can change the colors there. There you go, right? I mean, it looks kind of funky, like probably not the theme of all this, but just for illustration purposes, you can see that we can do that. Um, and so you can make all these adjustments, and right here, you can it this will put it on the different part of the screen, the different zones, so kind of standards there, and then you can drag these sliders to move them over left or right, and I'll just go back to zero. So then what you do is you go up here to this little plus sign right here, and you click it, then do create style, and then you can just name it uh whatever you want to, and then I just save it, and here you go. So it made those changes to our entire catch, our entire track, and now we have that is our whole thing by open AI, um claw, or one thing. I'm probably not talking clearly that's more the case, but and um, but yeah, so that's really good about that. Um, and so this is what I'll do when I start start, I'll implement that into here. And I'm not sure if I do that now, if it'll happen, but nonetheless, you know, I'm really happy with with what I've got so far. I'll continue working on this and um and just see how far we can get with it. I mean, ultimately, you know, you want to have something, you want to reduce the friction as much as much as possible and increase the efficiency. And uh for somebody like me, I'm not good at video editing at all. So um it's a daunting task. I have a lot of respect for good video editors. Um, it's a it's a science, it's an art. And so that's why using um these these tools now that we have our at our disposal, Claude, ChatGPT, Grok, whatever you're using, um, you can come up with well, the Grok API, but you know, it if you're using OpenClaw, we'll say the agentic programming and just using something like that I was using here with conductor, which by the way, that's the whole reason I started doing this, was just to try out this this um new conductor, um agentic IDE internal development environment. So it has everything you need. You know, this is various. If you've worked with uh cursor or Visual Studio Code, one of those, um, this is kind of like that, you know, it's just its own environment. You got you got a terminal over here. I haven't even I have not even gone and looked to see anything, I haven't even gone into settings. So um that's pretty impressive that it can get up and running and it almost got me there right up to the end. And uh, you know, if I was uh more patient with it, and I've you know, you can invoke skills too. So if I if I you see here with a slash command and then like debugging, and I didn't even run a debug from here, so um I did a test script, but it didn't fix it. But so I just jumped back over to Claude Code, and that's what I'm comfortable with um entirely. And so what I did is I just went and I um opened a terminal from from Claude Code or from that directory that I was in, just right-clicked um and on the folder, and then you can open a terminal right there. And um actually let me show you, just so you just so you know for sure. I want to go right here, and what is this thing? Oh okay, so that's and conductor, and then I'm under workspaces, and this is pod pro and that's Memphis. So this is the project itself, right? And so, like this is the back end, the server. And see, that's the.env, that those are our secrets and everything. And um and if you're not seeing these right here, that that's because they're hidden press uh on a Mac, press command shift, and then the period, and then it will it will make them come up. That's how you cycle that on and off. Command shift, period. And but so you go to the right here, and you can right-click it and go down to mind says services, but yours may say something different. But just just find this uh terminal at folder. I use iTerm2. You may want to download iTerm 2, just Google it and you can download it. It's a terminal uh shell that's really cool, works really good. I like it. Um, but at the end of the day, it's just a terminal. So you you'll have new terminal at folder. You open that, and you see here that it opens, it's going through, and it opened right here at the and this is the path right here. So conductor workspaces pod pro one, which is what I named that, and then Memphis. That's our actual project. And so now what I do is I run Claude and then dangerously skip commission or permissions here. This is the flag with the two hyphens, boom, boom, and then this is just one when you're putting, you know, this is like a compound word essentially. Press enter, and when I press enter, I should run Claude, what happened? That's weird. I'm going to control C and I'm gonna run it again. Oh see, it didn't, it didn't do that for some reason. That's weird. Claude There we go. Okay, so you see the first time it like for, I don't know why it froze, but I if you just press Ctrl C, uh, that will take you back out, and then I just ran it again. Claude dangerously skipped permissions, and now my Claude is open here. So that's really cool. So I really hope this helps, and I hope it helps. Um, you know, if you're interested in using this um conductor, I would I recommend it. I'm gonna I'm definitely gonna keep it on my computer, that's for sure. I want to be playing with it. This was the very first thing, uh, other than that game, I guess, that I showed earlier, uh, that cosmic thing. That was a one prompt. I mean, that's pretty amazing, and got it up and running and boom, opened it. It's like holy smokes, and it worked perfectly. So, this this uh Adobe extension for Premiere Pro is uh much more complicated, of course. And so it had a little bit of issues, but I, you know, being my first time, I'm sure a ton of that actually, I'll take a hundred percent of credit, or uh not credit, yeah. I'll take credit for it not being able to do that, you know, user error for sure on my part. Um, if I would have uh instructed it better, I'm sure I could have definitely given better prompts. My prompts aren't good. Uh, I need a lot, I have a lot to work on myself. Um, and also I just don't know the the program. Um you know, this is my first time in this. Um, I don't know what I guess it just imports all the um clawed these plugins. So, you know, I know I have like I know I use these and I have a lot on my Claude code in itself. So, you know, maybe these probably are you know imported from Claude Coder. I don't know if they're native, they may be native. I haven't read the docs, you know, so I don't know anything about this, to be honest. I just but that's the beauty of this, and man, like you don't need to um be the sharpest or brightest person, you just need to just play with it, just tinker with it. And um I really believe that if you just have enough patience and work through the errors that you know that we're really we're really in some cool time, so there's no reason to go out there and pay somebody a bunch of money to develop you a program that that you need for whether it's you know what at work or whatever, and uh because we're all fully capable of doing this ourselves now, and um you know if you want to go out and pay somebody to do that, that's fine, but you're probably not listening this far into this. So uh I definitely would suggest that that um you know try this stuff out if you haven't already. Uh and if you have, then you know, just keep rolling with it. And I think it's really cool, you know. Been playing around with this stuff now for oh well, about six years, I guess, and just trying to, you know, make it all make sense. But well, really chat GPT was my first uh back in March of February or March of 20 or January early 23. And um so immediately I saw the power of that, and I was working on the past few years. I had been working on different programs and um, you know, just knowing nothing. I'm not I I don't have any uh formal background in any of this stuff, so I'm I learned just like you, and um it was it was really nice to have chat GPT at that point, and like years of struggle just you know, all came to um, you know, it all paid off with Chat GPT, and um real quickly I saw how powerful it was, but at that point I had already been, you know, working with like uh blockchain stuff and solidity and things like that, and trying to build, you know, messing with websites and building websites and then WordPress, you know, but WordPress is just a um it's just a plug and play thing, anyways, and um it's really convenient, you know, WordPress is great, uh but now everything is so we're capable of doing all of this ourselves, and you know, I truly believe that if you just have a Claude code or a Claude subscription or or uh or a chat GPT and open AI subscription, get one of those subscriptions and start doing this stuff. Um that and then you get a GitHub account and you get a Cloudflare account. With those three things, a a chat, a Claude, GitHub, and Cloudflare, that's all you need to get started. And um, you know, to to build like real programs, like like back end stuff, like complicated real stuff. Like really to build real programs, all you need is just a claude or chat GPT subscription, open AI or whatever, you know, uh Gemini of Google, whatever, whatever, whatever you want to use. Uh, but get one. And with that, you can um start start playing with this stuff. And and the beauty of it all is we're using natural language, you know, and and so you can have a conversation back and forth with it and just tell it what you want and tell it what you have. So um tell it you have, you know, set up your GitHub account and then set up your Cloudflare account. Both of those are free to use. Very important. Both of those are free to use. GitHub's free to use, and Cloudflare is free to use. They both have paid tiers, of course, but the the free tiers are all you need to get started. And um uh maybe Cloudflare. You might have to, you know, pay 10, 20 bucks a month and some things on Cloudflare if they charge you for like ingress or egress fees. If you have data moving in and out, you know, um, there may be some charges there, but it's very minimal. Um, so I highly recommend Cloudflare. I could not recommend Cloudflare enough. And it's a great service. I've used Cloudflare for the past six years, and they have been nothing but great. They've been great. Uh their service has been fantastic, you know, uh as far as just their um what they do, the services that the you know, software as a service, the services that they provide. And also uh just as a company, it's a great company, you know, never any issues with Cloudflare. Um, you know, fair billing, they send you the state, you know, everything is clear, everything's crystal clear. There's no, oh, what's going on here? You feel like they're trying to upcharge you for anything. No, actually, with Cloudflare, you feel like you're kind of getting away with like stealing some stuff from them because they give you so much for free. So I highly recommend Cloudflare. And with those three things, so let's go over it chat GPT or Cloud Code, um, Cloudflare and GitHub, those three things get those three things uh started and then connected. So the way you get those connected is you start interacting with your uh Cloud Code agent or your Chat GPT agent, and you tell them that um, hey, I just opened this Cloudflare account. I need to um get you the agent, you know, uh Claude in this case. I need to get you connected to Cloudflare, and I want you to do it. And so it'll say, okay, but I need you, I can do that, but I need you to do X, Y, and Z first. You know, it may tell you to go to the Cloudflare account and it will tell you then it'll tell you go to setting or profile settings and go find this, or I need this API key. Here's how you do it. And if it doesn't tell you, if you have any questions, then you just ask it. And so it will surely be able to walk you through all of those processes with uh setting up uh Cloudflare and GitHub, getting getting them connected to your agent. And once you have them connected to your agent, then they are connected through your terminal, right? Where we where we communicate to our agent. Um, you know, you can you can use the the the GUI as a graphical graphics, you know, the the programs themselves like we're seeing here. And this is really cool, you know. I'll actually like uh you know, I like this setup. This is really nice. Um, but generally, you know, when I'm talking about it, like first, you know, I work really just in the in the terminal is where I stay, but I'm testing this out. But um, but so you you want to get those connected, and once you get the agents, once you get your uh Cloudflare and your GitHub connected to your agent, then they're basically permanently connected unless there's something that gets disconnected, but that's usually rare. And if it, you know, if that happens and you just go back to your agent and say, Hey, I need you to reconnect back to Cloudflare, whatever, or reauthenticate. Um so so yeah, man, that's about it. And it's really simple and it's really easy. And then once you get and once you get those up and running, then you can um, you know, you for your Cloudflare account, you store your source code on, and you can you want to store it private, you know, you can set it and you're when you create repositories or repos, you want to save them private, unless you want to make an open source thing, uh, which I you know if you build something cool, you should open source it. But if it's personal, you know, definitely make it private. And um, so that's your source code, and then that's where you do your git pulls uh from when you get started, and then push when you end. And so that keeps your source code fresh and up to date. And then in your cloud flare is where your actual program is running. So your Cloudflare in GitHub, you'll just have like it'll you'll have your all your uh code uploaded to there, but it will just be um it will just be static, it won't be doing doing anything. But on Cloudflare is actually where you implement the pages and the workers. Cloudflare is where you have the R2 databases, you can store uh they have stream where you can store videos, you can actually do live videos and now so you know if you want to do live podcasting or anything like that, you can you can run it through there. Um, you can store just regular MP4 videos, you can they have an image uh storage database, it's cheap, cheap, cheap. So uh once again, Cloudflare. I mean, they have email, they have uh domain, you know, DNS domain stuff. You can buy your domain on Cloudflare, you can you can do just about anything on Cloudflare. And you know, Cloudflare is a company that's always you know, always moving forward and um making positive changes. So it's they're they're always coming out with something new. And so I'd highly recommend it. So thanks for watching. And um, I want to continue building this, and you know, maybe next uh few days or so when I um get back to this, I'll I'll um I'll press record again and and we'll we'll see what we have at that point. I think I want to I need to get the transcriptions right, make sure our transcriptions go to the clips, um the captions, and then because that's very important for social media, right? You want the captions popping up because most people aren't even listening to the stuff that they're uh scrolling. And so once I get the captions right, that'll be good, and then I'll be able to build out um whatever else I want to build out. So I'll be thinking about this, and that's the thing too, right? Like we're now we're now not limited in our capabilities, we're limited in our imagination. And so if we can keep that in mind, we're not limited in our capabilities, we're limited to our imagination. And you know, so I'll be thinking about this, and the more I think about it, the more I'll think about things like oh, um, so let me uh if I click on this, right? Let me zoom in on the the timeline right here. If I click there, then I go to effect controls. You can I'll just do a quick little demo here about so I'll go to that's this is where the start of this clip begins.

SPEAKER_02 1:48:08

Yes today. You've probably heard about it, you've probably seen the little lobster with the claws, but you have no clue what anybody's talking about.

SPEAKER_03 1:48:16

Or maybe you're the person okay. So, like if I start right here.

SPEAKER_02 1:48:22

Yes, today.

SPEAKER_03 1:48:23

You've probably heard about it. Okay, so I'm gonna zoom in even more, and I'm gonna uh let's see, right there, if you click and drag, you can see the audio better. Yes today. Okay, so what I'm gonna do is I'm gonna stop right here before I go. And I'm going to press Command K and I just want to trim it. You've probably heard about it. You've probably seen the little lobster with the claws. Okay, and then press Command K again, right there where I stopped it. And now you see how it highlights this. If I so now the effects, so we're one, we're clicked on this, so that's what's we're seeing here in the effects controls, but also we're seeing this because this little timeline thing is over it. So that's why we're seeing this here. And so what we'll do is we can scale it up right here. You see this, and so we can, you know, just drag it. We'll just we'll just say we'll do it like right there. 135 is what it's set to. So watch this. So when I go back here to the beginning, the first part, this little clip is still at 100% because I cut it right there and then right here. So this is 100%, and then this is 100% over here. But here is set to 135. You see that? But it's not showing me the 135. Oops, I just right click something on accident. I don't even know what that was. Um, it's not showing me 135 because my timeline thing is. Right right here. So that's why I'm seeing that when even though I had this clip here that says 135. So watch this.

SPEAKER_02 1:50:13

You've probably heard about it. You've probably seen the little lobster with the claws.

SPEAKER_03 1:50:18

Okay. So you see how it zooms in, but watch, I can also add keyframes. And what keyframes do podcasts today. You've probably heard about it. You've probably seen the okay, right there when I do the little claws thing. I wanna right there. I'm going to go up here to scale. I want to click this little uh stopwatch thing, and you see how I added this keyframe, and then you can this is actually the same slider, but this is for just this little segment here. So you can even drag it over and say, I want to set a keyframe here, set another keyframe, and say, I want to take this down to 100% or 110%. And so now watch what it does. It's gonna you see how it zooms in and out and watch, it's showing you right here what's happening. So if I go back to the beginning, it's gonna play at 135. You've probably heard about it.

SPEAKER_02 1:51:26

You've probably seen the little lobster with the claws, but you have no clue what anybody's talking about.

SPEAKER_03 1:51:31

Or there you go. So there was uh I didn't even, I was looking for it, but that's just a subtle little, you know, it's going from 110 is the final keyframe. I have it set right here, and then this long the remaining segment of the clip is at 100%. So you see a little bit. You see that right there, okay. But so that's keyframes, very simple explanation of keyframes, and you can also um you know change um this here, and it should make it you've probably heard about it, you've probably seen the little lobster with the claws. Or do I have it backwards? Is that an ease out and an ease in? Let's see. You've probably heard about it, you've probably seen the little lobster with the claws. I think so. But yeah, so you can mess with that stuff, and it just changes just the slight approach. And if I were to open this up, you could see like the curves and stuff, or um, you know, it gets very complicated, and I'm not that good at this stuff, but this is so that this concept of keyframes is in everything, so it's also in position, right? So, like this. Um, I'll show you right here. Let's say I wanted to add a text, and I said today, you've probably heard about it.

SPEAKER_02 1:52:53

I'm gonna I'm just gonna say you've probably seen the little lobster with the claws, but you have no clue what anybody's talking about.

SPEAKER_03 1:53:02

Or maybe you're the person that does have but you've probably seen the little okay. Let's just do a silly little thing. Let's add let's add a text. So I'm gonna go up here right here and click text, and it's gonna add text to the screen. What is happening? Why did it do that? Click that and then click. I don't know why. I'm kinda lost on this here. I'm confused, I'm confused.

SPEAKER_01 1:53:53

Do I want the type tool? Let's see, it's unlocked, it's not locked.

SPEAKER_03 1:53:58

So let me try to oops, I didn't mean to do that. I'm not trying to do that. I'm trying to type. There we go. Um, I I guess I just had to toggle. Well, when I did toggle, it came up. So okay, so it's gonna put a clause. So I want to right there where I have it, finally got the type tool open, clicking this thing here, and um, if it doesn't come up, well, there isn't a video there, but I don't know. Nonetheless, I got it open. So I want to put you see how that pops up right there when I click. I want to say claws, oh claws. Used to type in uh clawed, claws, and of course, we could go over here and make all those changes, but we're not gonna do that. So I'm just doing it probably seen the little lobster with so I want to bring this down to where that clip is so you can trim it down, and so instead of just having it just pop up right there, what you can do is you can have it start over here by dragging this. You see that? Okay, now I'm gonna set my keyframe over there, and you see how it the little keyframe right there, you can barely tell. And then I'm gonna bring it over right here, wherever, you know, wherever I want it to. I want to bring it right here. I want to make it fast. And so the closer you are, then I want to set another keyframe. So now we have our two keyframes, and oh, that's scale. Wait a second, sorry, I was on the wrong one. Um that's okay, it doesn't matter. I'm gonna leave those the same. Actually, I need to put a keyframe on position. There we go. So, right there, and if you want to delete a keyframe, you right-click it, clear, and I'll delete that scale one. So we're just working with position right here. And um, so I have the keyframe there, and then you can come over right here, set another keyframe, and then drag it over. And then it will it will do fast. Watch this.

SPEAKER_02 1:56:12

The little lobster with the claws right now.

SPEAKER_03 1:56:16

If you want to make it the little lobster, you just drag this keyframe over. The little lobster with and once again you can do the I believe this is uh so it showed up differently, you know. Earlier it just said ease in, ease, ease out, but it's also under this one right here. You can go to ease out, and then I think this is supposed to be ease in, I think.

SPEAKER_02 1:56:43

Well, the little lobster wheel.

SPEAKER_03 1:56:45

Yeah, it looks a little better. You see that? It's just a little smoother.

SPEAKER_02 1:56:48

The little lobster.

SPEAKER_03 1:56:49

It wasn't so that first one was hurting.

SPEAKER_02 1:56:51

The little lobster with the claws. So the little lobster.

SPEAKER_03 1:56:54

That's pretty cool. And then if you want to, um, you can add a um you could you can make it go back off the screen if you wanted to. Um, so in that case, we would put a keyframe here to say, okay, we're right there, we're still at the same position. And then when we the clip ends, we're gonna set another keyframe and we're gonna drag it to the left. So that at the end it will be going out. So now we have a functioning little thing, uh text that slides in, and we'll show it from over here.

SPEAKER_02 1:57:35

You've probably seen the little lobster with the claws, but you have all right.

SPEAKER_03 1:57:40

That was maybe a little too fast exiting, so um, I'm gonna move it right there. The little lobster with the claws, but you have to actually I think you can um copy this. Let's see, I'm gonna copy it, and I'm going to clear these right here. I'm gonna backspace and delete them, and I'm gonna set that right here. And let's see if I can paste it. Oh, it just copied one. Okay, but oh, but the same thing, it would take it to the same um paste. Oops, nope, it didn't. So, yeah, I'm sure there's ways you can do that. But the little lobster with the claws, but you have yes, you see, I I was messing it up right there, but when I had it right here, it was working. There you go. So the little lobster with the claws. But keyframes are very important because keyframes are everything, and um and premiere pro and really a lot of other uh Adobe After Effects if you start using that too. And you can also, you know, like down here, opacity. Let's say instead of having it just go off real quick, we wanted it to kind of fade away. Well, there's two ways you can do that. You can go down here, and this is the easiest way, really. Just press Command D, and you'll see these cross-dissolves thing pop up. But you see, this front one did it popped up too, which will make it go, it will make it fade in, which isn't what I'm going for. I'm only trying to get it to fade out. So you can right-click there and go uh clear, and it will clear that so it will stay 100% opacity the whole time. Now, over here, you see it starting to fade away because the cross dissolve is there. But notice it starts fading away before it goes away. So I want it to start fading away right when it goes away, like right there. And so I'm gonna click and drag this cross dissolve over right there, so it'll just be a subtle fade out at the very end, right there. You see, it just kind of looks a little clear. So, yeah, just a few things you can do with keyframes. And um, so I'm going to try to make our program here, I guess we'll just call it Memphis, get Memphis to um do all of this. You know, eventually we want to have it to where we can um sit down, record our long form content, and then feed it into our Adobe Premiere Pro, go to one plugin or extension. I'm sorry, I keep calling it a plugin, but go to the this extension and have it do everything for us and then be able to not only you know make the make the clips as far as like it's already done and it's already intelligently made the clips. So that's really kind of the hardest part because now we'll be able to implement things that it can already do. You know, this stuff right here is just this is just hard-coded stuff that it will our agent will be able to um should be able to do. And what I need to focus on doing is I have a bad habit of trying to tell my agents to do everything at once, you know. Hey, I want to do all of this at once, which you know, realistically, they can get really far, but then you know, you kind of get like 95% there, and that last five percent you're trying to put the pieces together, and it can kind of become a cluster. So um I may what I'll probably do here is I'll probably just do like one thing at a time. You know, okay, we need to um like be able to add add text like this right here. This would be this would be a great thing. I'll probably start with this actually, and tell the agent, okay, we need to add contextual text with the type tool, I think is what this is called. Yeah, we want to look to make sure we know what it's called when we tell our agent. So with the type tool, you see that? And uh let's see if I get there, you go with the type tool, and we want it to be able to be contextual, you know. Um, and whatever the mood is, we you know, mood's very important, and this is something that I've started uh just just starting to study myself because I know nothing about you know film, like I know about like photography and like videography in the sense of the mechanics behind it, but I don't know the first thing about the editing part of okay, let's put this story together. We're trying to tell the story here, right? And so that's what I'm uh I'm starting to uh look into myself, and I need I'm gonna I've told myself I wanted to devote a lot more time to learning that because now this is the bottleneck, right? Like we're talking about the ability, like the bottleneck isn't me being able to build this program anymore. Because here, you know, a few months ago, I was um incapable of building a program like this up until all this, you know, the agents. And so we need to take full advantage of them and and build out okay. Maybe you're not interested in building out a podcast thing like this, but maybe you're interested in building out something else for yourself that would be that would be useful in this sense. And um, you know, get creative and just see what you can do. So appreciate you watching. Thank you.