Transcription: OpenAgents Episode 137 - Projects
June 9, 2026 ยท View on GitHub
Source: https://x.com/OpenAgentsInc/status/1852082485141573657 Wiki source: https://raw.githubusercontent.com/wiki/OpenAgentsInc/openagents/Video-Series.md Media title: OpenAgents - Episode 137: Projects Next we emulate Claude Projects. Their UI is g... Upload date: 20241031 Transcription model: gpt-4o-mini-transcribe Generated at: 2026-06-01T15:28:42Z
Machine-generated transcript. Review speaker labels and wording before using this as quote-grade source material.
[00:00] Speaker: Okay, you can switch between teams and your personal accounts, but now within a team, I want there to be multiple projects. And we're just going to be kind of emulating what Claude does with projects. So let's learn about how they've introduced Claude projects. Our vision for OpenAgents has always been to create AI systems that work alongside people and meaningfully enhance their workflows. Blah, blah, blah. Pro and team users can now organize their chats into projects, bringing together curated sets of knowledge and chat activity in one place with the ability to make their best chats with Claude viewable by teammates. Claude can enable idea generation, more strategic decision-making, etc. So one thing that I love about having products that are not from labs is that they're never going to use OpenAI's 01 models if they're good in Claude projects. And ditto for OpenAI. They'll never use Sonnet. I don't care. We're model-agnostic. We'll use Sonnet. We'll use 01. We'll use Open models. We'll use whatever is good and best and just have one interface that you don't need to go bounce between all these different labs. Isn't that nice? So when they pioneer cool UX like this, we're just going to kind of take it. So each project includes a 200K context window, the equivalent of a 500-page book, so users can add all of the relevant documents, code, and insights to enhance Claude's effectiveness. Yeah, and there's all sorts of opportunity for, like, doing more surgical RAG that I found even Claude's interface could be improved upon. Okay, so avoid the cold start problem. Projects allow you to ground Claude's outputs in your internal knowledge, be it style guides, code-based interviews, blah, blah, blah, additional context. It's just preventing you. Like, whereas right now I'm saying use these five files, you just don't need to keep doing that. Okay, custom instructions for each project. It's not, none of this is complicated. And then you've got artifacts. We may or may not do our own version of that for the code. We haven't really needed it because we have a direct connection with GitHub. But, you know, maybe we'll do artifacts at some point. But for now, just kind of getting the projects going. Claude team users, blah, blah, blah, shareable projects. Okay, cool. And then let's see if we can remember. I think it's for pro only and I don't have pro anymore, so I can't even look at the UI for it. But I'm going to guess that, let's see if they've got any, like, screenshots in here. So, let's see. Man, soon I want to add image input soon, so I can just, like, grab a screenshot of this stuff and drop it into open agents. So, yeah, they've just got, like, project knowledge on the right. I guess we could do that in, like, a right sidebar, maybe. All right, let's mock it up. We've got a doc for teams. Let's start a new branch here. Git checkout projects. Project structure. What is it? Oh, that's our project. OK, so we've got teams. And what I'm going to do is pop over to v2 and have it create us a document similar to this. Change this to projects. Look at boom. And why don't we also look at the test results? We've got a lot of broken tests now from the CRM stuff. That's OK. But give me the list of the tests so it knows what all is working and what's not. OK, look at docs slash teams and test results. Now we need to flesh out the UI and such for projects. So write a doc slash projects.md. Oh, and let's scrape that blog post that we were just at. First scrape this. Let's make sure we have that tool enabled. That's all good. Write it. Projects with how we'll make open agents projects. Very similar idea to the cloud projects minus artifacts and otherwise include info like the teams does for projects. Thank you, Firecrawl, powering our web scraper. Ooh, it stopped. Why did it stop? Oh, my gosh. It got broke. No, thank you, Firecrawl. Why you fuck up? Let's just try one more time. That might be a Vercel issue. Method not allowed. I have no fucking clue why it says that. Well, this is one of the reasons we've gotten off of Vercel is because they just get these weird ass errors. There's caching, just like. Here's info about it. I'll paste it in like an idiot. OK, that's all right. The tools that I need to pull over from V2 to V3 to work in V3 is largely the rewrite file and the scraper. I've got the viewing working. I don't really use the branches and issues. It's less important. Open Agents implements a project-based organization system that allows users to group related threads and resources together, enabling more effective collaboration and knowledge management. This document outlines how projects are implemented and used. OK, a project can belong to either a user, personal, or a team. Project service containers for related threads and resources, custom instructions and context, organizational units that contain chat threads, documents, and files, custom instructions, shared context, knowledge base, models, and relationships. Looks good. Let's just do it. Just do that. Just do it. OK, what's do I even have a project test? Yeah. OK, look at boom and see if anything is missing. Don't edit it yet, but suggest here if any tests should be added or changed to reflect the doc. Custom instructions. Cool. Context window limits. Custom settings. Tone, formal, technical writer. That's interesting. Add just, OK, so first of all, did my project test pass? Yeah. And then we don't yet support file uploads. So let's do add the tests there for number one. Context window limits. Project can have custom settings. That looks good. Knowledge base tests. Team access. Yeah, that's cool. User can view. I like that. Project validation. I like that. Threads inherit project context and instructions. That's cool. Yeah, let's do that. Not eight. Can be archived. Yeah, let's do archived. So, yeah, I think we're gonna be able to reverse engineer lots of interfaces like this, pulling in the majority of the, like, core functionality. Simplify it. Connect it to what else we're doing. And then just have one all in one. As I say in, you know, the vision for this V2 product, you know, described it as your all in one AI productivity dashboard. And then this ended up just kind of, we just focused this one really on developers. Like, we just never added other tools. Cause I didn't know what businesses would want until we got in connection with some. So this stuff has largely been kind of driven out privately, but now it's like, let's build it into the platform. Um, doo doo doo. All right, let's try it. So I'll say yes, go edit. Boom. Project factory and projects. So with one sort of master dashboard for your productivity, your business productivity, then it just comes down to, like, how many integrations and features that we have relative to the others. And if we combine this sort of like one beautiful interface at a trusted brand, open agents, then we open that up to plug in developers to contribute plugins in a marketplace and earn some revenue share for that. Um, we'll have by far the best experience and then we will laugh at all of these little vertical agent startups as we absorb them. Doo doo doo doo. I'm really excited to do the long-lived agents video. That'll probably be next week. I want agents working while I sleep. I want it. Blah, blah, blah, blah, blah. All right, we had seven failed tests and now we have two failed tests. Read boom and fix those tests. You'll need to add something to boom. And let's put that and write the relevant test for that in boom. One failed. User can view. OK. I don't know what that can thing is. If, you know, edit, boom, make it work. The can method is part of Laravel's authorization system. We need to find a policy for the project model and add the authorization methods to the user model. Let me fix that. Cool. Project tests pass. OK, so that's long enough for that chat. Let's add a new one. So let's say, um. Let's get all the project-related tests. Put that there. And then we'll say, read boom, read readme, test results. And let's have it read project test. Edit projects MD to add any other relevant info based on what we have for passing tests. Don't remove anything. And then next we'll do some kind of UI. Can we get a view of the what it's like to create a project? There we go.
[19:59] Speaker: All right, let's write this up. Let's say, Oct 31, okay, so for the create project screen, it should say, create a personal project if in personal context, or create a team name project. What else? Fields, what are you working on? What are you trying to achieve? Text input, text area, three rows, placeholder, name your project, describe your project's goals, subject, et cetera. Okay, let's just see what changes it makes here. Make sure that's comprehensive. Nice. Okay, so in a new chat, I'll say, open readme and docs projects, MD. We now need the create project screen. Look at routes web.php and this and this. So write a create project page modeled on the above, modeled on the team create page, and edit routes web to add the route. Okay. Eventually, it'll be awesome for agents to gather this kind of context by themselves. Create a project model. We already have a project model and migration. We do need a project controller. Look at this, projects, and project test. All right, what route is that at? Okay, it will be at projects slash create, which currently won't work because we don't have the controller. Not bad. Create a testo team project. Yes. You're gonna be making this stuff fast, people. And then I guess maybe we'll put project in the sidebar. Well, does it work? Is this my active database? I guess I should migrate the database and start fresh. Uh-oh. I think that's something we broke in the last episode, maybe. Oh. Fix. Clean that up later. Okay, I don't have a button for create project. Actually, you know what? Let's add one real quick. Let's do this. Edit boom to make a new project button under new chat. Also, change that post request for new chat to be a get request instead. Inertia link, not a. Let's see if it knows what I'm talking about. And then what should our icon for project be? We're using Lucide, so maybe like a, should be a folder? These look cool. Maybe folder root. Oh, look. Oh, it has a folder with a plus. Yes. I didn't even ask it. It read my mind. Oh my gosh. New chat, new project, new chat, new project. Okay. Okay, what happens if I do this? Oh my goodness. Ooh, did it make it? Yes, it did. You jibbidy-doobidy-doom. Okay, so we have a new project, but it just put me on chat here. Folder plus. Amazing. Amazing. Okay. And that means I should be able to fix this routes web back to get. I think there's a test that might expect it to be post, but we'll fix that there. Okay, good. All right, so we can create a new project. Now we need like a view project page. Although actually, I think first we'll do, kind of like there's a chats thing there. We'll do that for projects. Okay, so now in that same file, above chats, put projects, listing the user's projects or team's projects, if that's team is selected instead of personal. And then I think that logic is handled in here. Edit boom to pass projects in using similar logic from teams on what to show. Same collapsible group type. Let's see if that works. CRM, chats, projects. I want this interface. I want it for me. Okay? I want it. I want it. Oh, I already have nav projects. Don't keep retrying. No, I forgot. We already had nav projects as a component. Look at it and edit it and make the sidebar use it. Although I'm curious to see what that did. Yes. O-A-G-V. I don't know why that LI thing is there. Oh my gosh. Now I don't have a project page, so we'll do that next. So in their thing, when you create a project, that's what that looks like. I really want to feed an image in of that. Oh man. Oh, that feels good. Okay. All right, see, it links to projects one, but that page doesn't exist. Okay. All right, so lastly, let's do a little basic project show page. No, that's great. Now make us a basic projects show page. Edit routes web so at projects one, it shows the project. And then what do I want it to say? At top, it says name, project name, whether personal or team, private or shared. You can hard code private for now. Description, etc. Custom instructions and a list of knowledge files. For now, you can hard code something showing none. When in doubt, use the ChatGPT components. You can use any components from Boom to build that page and you might as well look at Boom for inspiration. Oh yeah. Oh shit. Goddammit. I had to overwrite your changes. Please make them all again. Well, I'm excited for agents to figure out all the version control stuff themselves so I never have to mess with Git again. That'll be nice.
[39:59] Speaker: Did it crap out? Rouse web, projects ID, let's see what happens. Page not found, show. God damn it. Oh, I thought I might have the contents here. Nope, that got fucked up, damn it. Well, let's test the recovery here. Look at routes web.php, docs projects.md, project controller. That's a good start. That's a good start. Let's create a new project called SuperTest demo. Nice. Okay, I think that's a good stopping point for now. See you soon.