Buzz Is a Bigger AI Moment Than OpenClaw. Here's My Bet

0:00I generally believe this is probably one

0:02of the biggest releases since open claw

0:05that changed how agents were used and

0:07harnesses were used across every single

0:10provider out there. And that is the

0:12open-sourced version of Slack by Jack

0:16Dorsey. Now, this project is called

0:19Buzz. And if I had to simplify it all

0:21the way down, it's effectively like

0:23creating your very own Slack, but also

0:28having agents natively inside of this

0:31kind of Slack application essentially.

0:34And I've been using it quite a bit over

0:37the past few days, and I wanted to go

0:38over start to finish how I've set mine

0:40up with this new project that I've just

0:42created for the purpose of in this

0:43video.

0:44So, first of all, the key thing is to go

0:47ahead and go to Buzz. I'll link the

0:49website below so you can just click on

0:51it and download it.

0:52But what I really really like about it

0:54is just the interactive ability with

0:57agents and how you can set different

0:59agents up inside of here while feeling

1:01very familiar because obviously we're

1:03also used to Slack. And of course, it's

1:05all open source, which is awesome.

1:07So, first things first, let's just cover

1:09the Slack and components. It's very very

1:12similar to everything that you would

1:13have seen and know inside of Slack. I

1:15can create new channels. I can create

1:17new forums. I can do direct messages

1:20inside of these channels. I can actually

1:22view who's in there. I can create

1:24huddles. Everything that you would

1:26expect when it comes to just having a

1:28standard messaging application like

1:30Slack.

1:31So, that's the basics kind of covered,

1:33right? You can do everything that Slack

1:35can do essentially. I can also create

1:37the ability to push data in via webhooks

1:39and and stuff like that. So, I can

1:41connect it into other applications and

1:43systems should I wish to do that.

1:46The key thing here though is obviously

1:47because it's open source, you can

1:48self-host it on a computer. Um and the

1:51really really cool thing about this is

1:54the fact that you can actually start to

1:55create and so, straight out of the hood

1:58you basically get three different agents

2:00that come with it. These are kind of

2:01inbuilt buzz agents Fizz, Honey and

2:05Bumble here. And you can actually see

2:07when I first arrived into this channel,

2:10you can actually see that they started

2:11messaging me as a team. And the really

2:14cool thing about how this is set up is

2:16agents will chat to agents really

2:18natively and instinctively inside of

2:21this application. So when you're

2:22chatting to them, you can say, you know,

2:24like uh hi Fizz, can you tell Honey to

2:29go and do some research on who

2:33triggerfi.io is.

2:35And they will go ahead and chat to each

2:38other and work with each other until

2:39they come back and do something. So it's

2:41really really cool for like agent

2:43collaboration and a place for you to

2:45actually see what is going on then. You

2:47can see straight away that they have

2:49both reacted to this.

2:51And that should shoot me a message back

2:53shortly.

2:54Now

2:55there we go. It's going to go ahead do

2:58this.

2:59So one of the cool things is you can

3:00actually create your own agent. So in

3:03theory you can actually go ahead and

3:04connect your Hermes into this. If you're

3:06still using open gloves, you can do

3:07that. If you have any other custom

3:09agents that you've created, you can do

3:11that through here as well.

3:14But what you can do is we can click

3:15create a new agent and then we can

3:17create them from scratch.

3:19So this is where you could title it. So

3:21for example, I could title this just a

3:22Claude code and we're going to go ahead

3:25and literally just use the Claude code

3:27Hermes. And this is really cool because

3:29what this does is this actually taps

3:31into your local Claude code

3:34subscription. So if you have the max

3:35subscription or whatever it is, it will

3:38tap into that. It's the same thing with

3:40your code x subscription as well. It

3:43does the exact same process.

3:45When it comes to using your Hermes, it's

3:47a slightly different setup there that

3:49you have to go about and you just have

3:51to connect it through the Hermes

3:52configuration there as well if you

3:54wanted to.

3:55Then we can say who can talk to this

3:56agent. You can decide whether it's just

3:58yourself or you, and you can go ahead

4:01and gate it. We can use a custom model

4:04here as well, so you can kind of preset

4:06it to being, you know, only using Opus

4:094.5, which is what we can see I've done

4:11over here.

4:12Now, the other cool thing is you can

4:14actually start to give it some specific

4:16instructions. So, you can say, "Okay,

4:17hey, you are the master of all agents.

4:21You will go ahead and delegate your work

4:23to other agents." And this could

4:25effectively become your chief of staff.

4:27And so, you can see where this is going.

4:29We can start to create a lot of

4:31different agents for a lot of different

4:34tasks.

4:35Then from that point on, what they're

4:37going to do is they're going to then

4:38start to work together and work in kind

4:41of collaboration with each other in

4:43here.

4:44So, you, for example, we could go ahead

4:46and let's say, you know, this we're

4:48we're running this internally. I could

4:50create a a bug channel here.

4:52And any bugs that come in here, I then

4:55could have my agents that run. The

4:56really cool thing is obviously because

4:58it's built on top of Cloud Code,

5:00it has access to all of my repos, all of

5:02the tools that I have access to control

5:05to. So, if I actually just click into

5:07the agent here, we can see firstly all

5:10of the different MCP servers it has

5:11access to, GTM Grid, that's the tool

5:13that we use to build all of our outbound

5:15campaigns here at Triggerfy, and A10

5:17obviously I've done a lot of work on

5:18there. So, go see the Hermes agent that

5:20it has access to, so on and so on. And

5:22we can see obviously all of the other

5:23stuff that it's connected to. So, it

5:25still has access to all of these tools

5:27that you connect natively in your cool

5:29code app or in your Codex app, and it

5:31still has the ability to use that query

5:33there. It just centralizes and creates a

5:36place to combine all of your agents

5:38under one.

5:40Now, the other thing that is really cool

5:42with this is we could actually go ahead

5:44and select the Buzz agent here.

5:47Here we could pick any specific that you

5:49would want. But one of the really cool

5:51one is this Buzz shared compute. Now,

5:54what this basically means is the if I go

5:57to the settings here is anyone who

6:01becomes part of this project or this

6:04workspace I should say, has the option

6:06to share compute

6:08across with other individuals. So, think

6:11of it like you maybe have a team of 150,

6:13you know, whatever it is. If you're all

6:15on on Buzz, you all have the option to

6:18share compute across this workspace.

6:21And the more compute that you share, the

6:23bigger the local model that you can use.

6:26Which basically means you can start to

6:28run local models for free as the compute

6:31gets bigger. Obviously, you can start

6:32with things small things like general

6:34four or queen or whatever you're wanting

6:36to do. But again, the more people that

6:39you add to this, the bigger the local

6:41model that you can run. And I really

6:43think that this is where the future is.

6:46Which is combining compute to allow

6:48people to run, you know, rather large

6:51models, perhaps something like a Kimmy

6:53K3 or a GLM 5.2 that is pretty big

6:57across multiple different laptops or

7:00workspaces or

7:01you know, whatever you're using and to

7:04have a free LLM running in the

7:06background.

7:07Very very very cool and I believe, you

7:10know, we're just at the start of looking

7:11at this type of technology and I think

7:14this will start to evolve now this is

7:16kind of come into fruition.

7:19The other thing is obviously we have a

7:20mobile app as as well. We can customize

7:22everything can be, you know, great. Then

7:25we have the workflow capability inside

7:27of here. So, here what you can say is

7:29okay, well, anytime a message is posted

7:31or a diff is posted and which is a code,

7:33a diff, a web hooks come in

7:36a schedule. We could run this like every

7:39day. We can then actually go ahead and

7:41define the different steps that it needs

7:43to do. So, we could say, "Okay, step

7:45one, the action is to send a D then we

7:48can obviously defined, you know, who

7:50that DM would be to, etc. etc. etc." And

7:54you can actually get your agents to set

7:55all of this up as well. So, they can do

7:57some level of kind of orchestration

7:59inside of your Buzz uh platform as well.

8:04So, yeah, that is basically the way that

8:05this works. So, we could say, "Okay, you

8:07know, hey, uh Claude code, can you work

8:10with Codex to go and review any of the

8:15latest PRs we have inside of our

8:20Triggerfy GitHub repo."

8:23Um and then a way that it's going to go.

8:25It's going to work together as a team

8:27with the uh kind of two agents that I've

8:29basically tagged inside of here. So,

8:31again, a really cool way of kind of

8:32combining agents. Now, I don't think

8:34we're at the stage of doing long heavy

8:37coding tasks at this point. I think that

8:39might come. This is obviously just super

8:41early days. It's just the first

8:43iteration.

8:44At the moment what I found is really

8:46good for little small fixes and little

8:48triaging work, perhaps little just kind

8:50of feature changes, UI changes, things

8:53that are small. If you want heavier kind

8:55of more involved coding tasks, still the

8:57terminal or the native app seem to be a

8:59better experience there. But like I

9:01said, I think this is going to be the

9:03future of how kind of work collaboration

9:05works and combining multiple different

9:07agents into one.

9:10Works nice.