Omnigent: Multi-Agent Orchestration & Real-Time Collaboration

0:04All right, Sunny. So, let's talk a bit about Omnigen. I saw the release. What is this about? Like, how is it working? Tell me Tell me >> All right, let me tell you. So, Omnigen is our new open-source project. We just released it last week, actually. So, it's hot off the press. So, I can tell you a bit about it, actually. So, Omnigen, I think there's like three things you need to know is that we're trying to solve the problem of control, composition, and collaboration.

0:29And the way we're doing that is we're building a new meta harness that sits on top of all all of the other coding harnesses out there. >> So, it's not like something It's not like third-party codex. It's like something on top of that. >> Exactly. >> So, you use one of these when you use Omnigen. >> Exactly. And that's why we call it a meta harness. Yep. And you want to see a demo? >> Yeah, let's let's have a Do you want to

0:51Do you want to come present? >> So, this is our documentation website, but let's jump over to our the actual application. So, over here, we have our application. And as you can see right over here, we have the ability to select any harness we want. HotGo, Codex, we even have Polly and Dedny, which are custom agents that we have built on

1:15using Omnigen. >> What these agents do run on from Omnigen? >> Exactly. Yeah, so these are our custom agents. And what it actually allows us to do So, for example, Polly what Polly can do is that when we start an agent chat with Polly, can we ask it to do a task? It can actually kick off a bunch of sub-agents to do

1:37sub-tasks for us. For example, it can tell uh Hot to do a certain task, and then it can tell Codex to do another task. >> for instance. >> Exactly. Yeah, so you have a really big codebase, and you have a lot of sub-tasks you want to do, you can spin up a bunch of agents that are all orchestrated with a parent agent called Polly. And it works somewhere is it

1:59running? When it when I run something, can I close my LTAP and it's going to run somewhere? >> Yes, but it really depends, right? So, if you have your machine as a host, you actually have to keep your machine running. Yeah. But, what you can actually do is have a remote server. So, if you have a remote Omnigen server, you can now run your machine as the host. >> So, you type the question like this.

2:21>> Exactly. >> In your LTAP. Mhm. And then it runs on the on the host. >> It actually runs on your machine. So, your machine is the host. >> Okay. Okay. Okay. >> Right? And then now you have an Omnigen server. And what you can think of it as is kind of a sync server. So, now let's say you connect from another machine. >> Yeah. >> It could be right here or you could be

2:43across the world. >> Like in my mobile? >> Yes, your mobile. You can connect to the same session and see the same session that you see on your local host machine. >> Got it. Got it. >> Yeah. Yeah. So, that is So, the collaboration piece of it, right? So, I think we are talking about collaboration already. >> Well, you can So, it means I can send you the conversation and you can load it

3:05from your LTAP. >> Exactly. Okay. Nice. So, actually I'll show you real quick. Let's just say we had a conversation going, right? So, I'm just going to pull up the conversation we had going where we built the Databricks app. On the top right right here, you can click share. And this is where I can put in your email. >> Okay. >> I can give you read or edit access and

3:27then grant you the access. And once I grant you access, I copy this link. I can send it to you. You can open it up on your mobile phone. >> I will see everything. >> Exactly. And not only can you see everything, if you have edit access, you have the ability to prompt the agent as well. So, we're both in conversation with the same engine and the same window. >> So, you see you see the agent working.

3:49>> Is that just in real time? Uh so, uh like I said, if talked about collaboration. >> Yeah. >> Um composition is being able to select your agents, right? >> Yes. >> And then now the third piece is control. So, one thing that is a problem, for example, is that all these uh harnesses out there, >> Sure. >> they lack uh certain controls. For

4:11example, if we want to uh limit your how many tokens we are spending, >> Yep. >> we don't have a way to do that right now. But in Omnigen, we do. So, if we look back into the UI, top right again, we have this icon right here with the ability to add policies. So, we already have a policy right here. So, I'm just going to remove it real quick so we can show what's going on. We have no policies to look at. And we now

4:34have this window where we can add some policies. >> Nice. >> Uh we have a bunch of pre-built policies already in uh built into Omnigen. But a good one that we always uh look at is this session cost budget policy. And this allows you to set a cost budget for your tokens. So, let's just say, for example, we put in a $10 limit. >> I know.

4:56>> We add the policy. The policy is now added to the session. So, now, right now we're at uh just a a dollar and 40 cents uh token cost. But let's say we keep prompting agent and we get to our $10 limit, the agent's going to uh stop. >> It'll stop and save. >> Exactly. And when it stops, it we have the ability to configure it, right?

5:18You can tell it to stop using high-cost agents. For example, Opus, maybe uh >> See that? >> Yeah. >> And then >> it will actually now fall back to using a lower-cost model instead. For example, Haiku. >> That's nice. >> Yes. >> Well, and so it's all open source. You can check it online. >> You can check it online. It's more versus >> from the agent.ai.

5:40>> Omni Jets >> I'm sorry. Sorry. >> Omni Jets >> Omni agent.ai. >> Omni agent. >> And so >> Go back in website right here. >> Right here. 1% open source, you can try it out today. >> Yep. All right. >> In lake cow, rise [music] data seeds, lake flow streams the flow. Iceberg and Delta lake, tables of glow. Unity catalog governs lake base OLTP strong.

6:04Genie takes to SQL MOSAIC AND KHIPU LONG. You simply [music] incohost like the fire. Data and dream scale higher and higher next gen. Drop it.