Transcription: OpenAgents Episode 119 - v2 Beta Launch

Source: https://x.com/OpenAgentsInc/status/1823109640357339628 Wiki source: https://raw.githubusercontent.com/wiki/OpenAgentsInc/openagents/Video-Series.md Media title: OpenAgents - Episode 119: v2 Beta Launch We demo our new v2 chat & auto-codin... Upload date: 20240812 Transcription model: gpt-4o-mini-transcribe Generated at: 2026-06-01T17:10:27Z

Machine-generated transcript. Review speaker labels and wording before using this as quote-grade source material.

[00:00] Speaker: Hey, we're live. OpenAgents.com. Let's do it. So you can log in with GitHub, Twitter, or email. If you're gonna be doing GitHub stuff, you should log in with GitHub. But if you don't, you can do manage account up here and then just connect it later. We're using Clark for auth, which is super cool. Okay, so we got panes, HUD panes. All of them are draggable, resizable, so you can configure your chat workspace how you want. If you command click or whatever the Windows equivalent is, you can open multiple panes here. Multiple chats at one time. Pretty cool. Okay, oh, that'll get increasingly cool as we add different types of panes like multiplayer and stuff, data visualization, et cetera. So models, tools, GitHub. Down here, you'll start off with free unlimited-ish access to GPT-4o mini. So you can try that out. And once you get it working, you can buy some credits, which I'm currently going to give myself some credits here, to use the advanced models. For the next few days, if you DM OpenAgents Inc, we will give you $10 of credits to play around with. All right, so I'm gonna use Sonnet. And that is the right repo. And then these are my usual four tools for my day-to-day workflow, all connecting to the GitHub repo, being able to create and rewrite files directly to a GitHub branch, and then view file view hierarchy. Okay, so I'm showing you what I do. So you can kind of see how this works. I'll give it a little briefing. And you can just chat with it as normal. I'm gonna give it some kind of starting paths to look at first, and it can traverse any of those and add those files to the context. If you're using one of the GPT-4o class models, it'll be able to do these kinds of tool calls in parallel. That's pretty nice. Claude just kind of steps through methodically. So one thing to be aware of is that all of these files right now, it's adding the entire context to the chat, which can be good, but also it can get expensive. And as you can see, my credits are deducting up here. So you will have to be mindful of your credits. And then feel free to switch over to GPT-4o mini. You'll also wanna have conversations be relatively short because the longer it is, the more the context that gets passed in. Okay, so it went through, went ahead and rewrote a file. Eventually we'll do more kind of like pull request approval, manual approval things. This is kind of errors on the side of getting stuff done quickly. So we got to pull in the better diff viewer, but you can see here, toggling between the older and the new, that's the older content. You know, we removed the hard-coded repo and it's nice also that the new content bumped up the storage key. I didn't ask you to do that. Clear out the old version. So we're basically making it so that this GitHub button down here doesn't just show hard-coded data or it doesn't show for everybody. It only shows for people who have connected to GitHub account. And if you don't have an account connected, it'll prompt you with like a modal to connect and what the instructions are. Okay, and it edited input settings. This definitely took longer. It takes time proportional to the length of the file because it's doing whole file edits. And then over time, we'll get that more optimized to use more surgical edits. So that's just been a forcing function for me to like write smaller files that fit better into context window. Okay, let's see what it did. These changes implement the requested features. We've removed the hard-coded demo open agents repo from the store. We've updated the input settings component to use the use user hook from Clerk instead of use auth. I don't know if that will end up having been correct, but we'll see. Add a state variable. Is GitHub connected? Okay. Blah, blah, blah. Use an async function to use effect hook. That's good. And yeah, sometimes it'll like get confused about client server components as well as like things like the Clerk docs. I don't think it's fully represented in the training set. So sometimes you've got to kind of manually pay some stuff. That's why I added some additional context there. But now let's take a look at it. The goal is eventually for you to not have to go over to the GitHub repo interface at all. But, so it made two commits. And it first removed those lines in repo store. That looks great. And then I've had to record this video a previous time because this would blow up the app if it did something wrong and tried to use the server-based logic in a client component. But it looks like this is pretty good. Putting the server action thing in a use effect. I don't know if this use user syntax is correct. Even though I copy pasted from the Clerk website, their React docs are not the same as their Next.js. Okay, let's try a moment of truth. Git pull. I have to worry about this blowing up the app if it fails. You don't. Because you won't be working on this app. It looks like it worked. At least it works like it didn't blow things up. So what happened exactly? Well, we need to test this out if we are logged out. So why don't we maybe log in with a different thing. Let me switch over to this account. No, it does not have an account. All right, let's log in with Twitter. That's weird. This might still have it connected. Dang it. Let me do this. Can I disconnect this? Okay, let me try this again. Log in with Twitter. Now that I don't have the GitHub account connected. Okay, look, we got a button. What do you think happens when I hit the button? Exactly what I wanted it to do to use the repo selector. You need to connect your GitHub account. Click your profile top left, blah, blah, blah. That is what I wanted. Okay, let's do it. Manage account. Connect. GitHub. This email address already. Okay, it said that previously, but it did it anyways. All right, let's test it out, I guess. Let's test it out with mini. And I'm gonna say, go to main branch. This is a private repo, so it won't work if it doesn't have access. Read everything in the docs and knowledge folders and summarize and advise on next steps. Please work. Be really cool if this worked. Failed to view. Okay, so I'm guessing, let me try refreshing the page. Let's try it again. Read everything in docs, knowledge, and summarize here. Hmm. Okay, well, I think that's just because there's a conflict there. I don't have another GitHub account that I can test this with. Okay, well, guess what? I'm gonna test this offline. And if you're watching this video with this in it, it means it succeeded. So please go try this out at openagents.com. Try it out for free with this. DM OpenAgents Inc on Twitter. We'll give you $10 of credits to try out the advanced models. Find a workflow that works for you. Once you get good and this is like you're getting value from this, feel free to add some credits there to keep using it. And please give us feedback. Tell us what works, what doesn't. We're gonna be making edits to this very rapidly, multiple times per week, particularly because we can use it to build itself. So please let us know how to make this more awesome for you. And we'll see you soon.