Transcription: OpenAgents Episode 038 - Agent Node Graphs
June 9, 2026 ยท View on GitHub
Source: https://twitter.com/OpenAgentsInc/status/1741887869055119630 Wiki source: https://raw.githubusercontent.com/wiki/OpenAgentsInc/openagents/Video-Series.md Media title: OpenAgents - Episode 038: Agent Node Graphs We introduce AgentGraph, our custom GU... Upload date: 20240101 Transcription model: mlx-whisper/mlx-community/whisper-tiny Generated at: 2026-06-01T20:26:28Z
Machine-generated transcript. Review speaker labels and wording before using this as quote-grade source material.
[00:00] Christopher David: Happy New Year. Let's talk about Agent Node graphs. So the goal for creating this
[00:07] Christopher David: agentic behavior is that we're able to see every single step that the agent takes.
[00:12] Christopher David: In a nice clean interface along with what the input and the output of each step is
[00:18] Christopher David: and then be able to kind of drag and rearrange these if we want to add new ones
[00:27] Christopher David: and be able to separately reason about each of these things. So this is an example
[00:34] Christopher David: of the agent that's going to power the initial chat experience here. So right now this is just
[00:42] Christopher David: making that call to the what is it that together API built previously. But it's about to have
[00:55] Christopher David: this flow here. So instead of just making the API call we wanted to do four things.
[01:01] Christopher David: We want to validate that the input is a valid chat message. So it's category validation. We want to
[01:06] Christopher David: convert the input to a vector betting betting. Compare the input to the knowledge base similarity search
[01:12] Christopher David: and then call the inference. This is just a basic rag flow as they call it retrievalogent
[01:18] Christopher David: generation and in the earlier videos you'll remember we kind of built our own versions of the
[01:22] Christopher David: embedding and the similarity search. That's just the like necessary components for RAG and
[01:30] Christopher David: making the API calls to any API call we could do open AI if we wanted to. We tried to use the open
[01:37] Christopher David: models via together and soon our open compute network for that as well as validation. So these
[01:43] Christopher David: kind of four categories are the four main ones. You can imagine that there will be another type called
[01:47] Christopher David: tools pretty soon. But so this whole idea of this kind of like library for building
[01:54] Christopher David: agentic workflows like it's something that we need and also I think it's something that is
[02:00] Christopher David: other projects will be interested in that as a standalone package and pulling that in.
[02:07] Christopher David: So for now we're building this within our own repo and we're just going to have it live in the
[02:15] Christopher David: agent graph folder in the root but this will be able to be extracted to its own separate module later on.
[02:24] Christopher David: We're inspired by these two projects like graph.js. This is something that I think comfy UI uses
[02:34] Christopher David: decent and then levers where I'm pulling most of the code from
[02:37] Christopher David: the panel for the panel they only have it for one GUI panel and I'm kind of multiplying that.
[02:45] Christopher David: But levers from the amazing point of mind rest developer collective behind re-expring and re-actory
[02:50] Christopher David: fiber and just use gesture. We're going to be using there stuff for a bunch of other things I think
[02:55] Christopher David: also. So it's nice to kind of pull code from them and learn from them because they have one of the
[03:01] Christopher David: there are some of the just top react experts that I've come across learn a lot from going through
[03:06] Christopher David: so we're pulling that code over. I've started that already in the agent graph folder here.
[03:12] Christopher David: And then have separately deployed a storybook. So a storybook for the so don't know is one of the best
[03:22] Christopher David: component libraries front end workshop for UI development. A front end workshop for building
[03:29] Christopher David: UI components and pages in isolation. So this kind of goes along with a philosophy called
[03:38] Christopher David: component driven user interfaces component driven development and we're going to try this.
[03:46] Christopher David: The idea is to build one component at a time like an avatar button or input or tool tab then combine
[03:52] Christopher David: the components composed small components together to unlock new features while gradually increasing
[03:56] Christopher David: complexity, warm header list table and then assemble pages. And this is kind of like the front end
[04:05] Christopher David: version maybe of what we had done previously with the test driven development like we have
[04:09] Christopher David: still a bunch of tests that are working fine for the like back end functionality. But for things
[04:15] Christopher David: where they're very front end heavy where we might want like you know a little widget to
[04:22] Christopher David: drag to connect one to the other being able to develop components in isolation is very helpful.
[04:31] Christopher David: So we have the storybook set up in the repo and it also now every time we push to the main
[04:38] Christopher David: branch of the repo that will auto deploy to storybook.openagents.com. So this is a publicly accessible link
[04:46] Christopher David: anybody can go to will put some introductory material there. And then we have stories for two
[04:54] Christopher David: different components the canvas and the node and there's some like auto documentation here.
[05:01] Christopher David: Make that a little bit bigger. But you can see for example like if we're testing the canvas component
[05:06] Christopher David: there's an empty canvas component and then if we make a change it'll like instantly show up there
[05:11] Christopher David: and then a canvas with one node and then canvas with four nodes and these four nodes are kind of using the
[05:17] Christopher David: demo data dummy data that is also the data that will power the concierge agent.
[05:26] Christopher David: So lastly let's talk about the concierge agent. So we've got this updated roadmap now
[05:31] Christopher David: with what we're launching this month. So we're we're building up to our auto dev agent
[05:39] Christopher David: previously called fairie now called auto dev and that's like the core functionality that we think people
[05:44] Christopher David: will happily pay for is like an automated agent that can live in your GitHub repo or any
[05:49] Christopher David: GitHub repo. Auto code, comment on issues, make a midsmit pull requests, all using open model
[05:54] Christopher David: open compute and you know paying people who are helping to improve the algorithm and such.
[06:00] Christopher David: But we're going to start with this concierge agent and that is what's going to live pretty soon
[06:05] Christopher David: at the home page of openagents.com or places landing page with the actual chat.
[06:12] Christopher David: That's kind of what we had mocked up there. This is just using the very basic API call but we're going to
[06:17] Christopher David: be probably offline next or so connecting it to use this new flow and you can see how now I've got the
[06:27] Christopher David: this used to look you know more basic but as I built these components in isolation yesterday
[06:32] Christopher David: it was very easy then when I was ready to upgrade them to just come over here and just drop
[06:36] Christopher David: into new components and know that they would all work together because I already built them in isolation
[06:41] Christopher David: to work together testing via the storybook. Okay so we're going to be doing some version of allowing
[06:54] Christopher David: people to edit the concierge based for their own use, probably not for other people's use
[06:58] Christopher David: because I'm with the concierge to be kind of opinionated on you know it's going to be
[07:03] Christopher David: deriving insights or regurgitating facts based on the knowledge base that we give it
[07:09] Christopher David: that'll be a nice good kind of example of what a knowledge base looks like but I wanted to be a
[07:14] Christopher David: you know introduction to the openagents product product so similar as if you could where to go to
[07:19] Christopher David: pi and I just like pops up this chat thing where you can ask your questions you can see I kind of
[07:26] Christopher David: stole some UI elements or layout from this yeah what you'd be able to ask it questions and then
[07:35] Christopher David: similar also to how with openagic you can kind of build up an agent via just talking to their GPT
[07:40] Christopher David: thing kind of want some of those same ideas represented on the openagents site so you can start to
[07:46] Christopher David: kind of maybe talk with the concierge to get oriented and start to build your own agent but for any
[07:52] Christopher David: agent I want to be able to be able to click over here and see exactly what's powering see these cards
[07:57] Christopher David: and then be able to edit them then we can do interesting things about letting ones that you
[08:06] Christopher David: edit then be listed in some sort of marketplace if you're kind of improving the algorithm of
[08:13] Christopher David: rag because you put in a particular algorithm that then gets listed and anytime anybody uses that
[08:19] Christopher David: then like some percentage of the payment for that agent to work flow goes to you building kind of
[08:25] Christopher David: community based revenue sharing and discover ability of each other's agent widgets
[08:31] Christopher David: all in a super open source transparent process is kind of the idea here okay so agent graph you can
[08:39] Christopher David: follow along there we'll keep this wiki more or less updated and stay tuned we'll have some
[08:46] Christopher David: bounties pretty soon for kind of adding and editing to the theme and such but check out storybook.openagents.com
[08:52] Christopher David: and click around with the notes see you later