Talking to Your AI Agents

Jack Dorsey, the creator of Square, has relaunched a new software called Buzz. It’s a group chatting platform built for teams and for AI agents of all sizes. I’m going to walk you through all the features: how to talk to individual agents, jump on a live huddle call, have multiple agents collaborate on the same Git repository project, and manage every single agent from one single platform.

To get started I navigated to the Buzz repository. If you want to install it, the easiest way is on your local machine. So I popped open my terminal, told Claude to install the repo (I already had the latest Buzz installed, so everything was good), and then you need to set up your local relay so it can connect to localhost and talk to you in real time. Once you pick the first option and continue, you are ready to open it.

Right off the bat you see the option to create a new identity key. What is that? Buzz doesn’t use passwords. They use identity keys to represent you. No password. You lose your key, you lose your access. So I created my unique identity key, copied it, saved it somewhere safe (really important), and clicked next.

After that you set up your agent harness. You can choose Claude Code, Goose, or even Buzz’s own harness. I went with Claude Code, hit install, then picked the Opus model for my default (feel free to use any model you like). Then I created my community from scratch on the Buzz website, signed in, and went back to the app.

Now I built my profile: gave it the username Eric, added a profile picture. Buzz immediately hands you three starter agents, Fizz, Honey, and Bumble, to help you get rolling. I clicked “Take me to Buzz,” allowed microphone access, and there they were.

Once inside, you see a welcome channel, a welcome-to-everyone channel, and a general channel. But the real action is in the agents section. There you see all your agents, with green status bars showing they are running. You can click any agent and choose “message” to start a direct chat.

I clicked Honey and said, “Hi.” You’ll notice a little “I” icon appear while the agent crafts its response. A moment later Honey replied. That’s the basic one-on-one interaction, and it feels snappy.

But you can also bring multiple agents into the same conversation. In the general channel, I referenced Fizz, Honey, and Bumble all at once and typed, “Hello, can you introduce yourself?” Every agent jumped in, each with a different personality: Honey, your writing and thinking partner; another agent as a researcher; the third more like an orchestrator. They all responded in the same channel, which is exactly how you want to collaborate with multiple AI helpers in one place.

Jumping on a Huddle

The huddle feature works a lot like a Slack huddle, except here you add an AI agent to the call. Up top, you click the huddle button, select the agent (I added Honey), and suddenly you’re on a live voice call.

I asked Honey, “Can you tell me a funny joke?” and it replied, “Man walks into a library and asks for books about paranoia. The librarian whispers, ‘They are right behind you.’ Since you’ve got me in a mood, I told my wife she should embrace her mistake.” Cheesy, but it works. The whole conversation gets transcribed inside the message thread, so you can scroll back and read every word. That means voice interactions are not ephemeral: they’re captured just like text messages.

You can also stop a running agent, create your own agents from scratch, or discover agents that your team has shared, bundling them into new teams for different projects.

Agent Memory: They Actually Remember Stuff

Every single agent in Buzz has a memory section. Anything an agent picks up about you gets stored there, and you can peek at it anytime.

Here’s an example. I jumped into the general channel and told Bumble, “Hey, I like pizza. I want you to remember that.” Bumble logged it into core memory. If you open Bumble’s profile and go to the memories tab, you see an entry: “Eric likes pizza.” No fluff, just a clear fact about the user. That same persistent memory is available for every agent, so if you teach Fizz something about your project or preferences, it sticks. It’s a simple but powerful way to let agents build context over time.

Pairing Your Mobile Device

You don’t have to be stuck at your desk. Buzz lets you pair your mobile phone with your desktop app so you can interact with agents on the go.

From the settings, I clicked “start pairing” and copied the pairing code from my desktop. On my Android phone, I pasted the code, clicked connect, and the devices matched. Instantly my mobile app showed all the same agents, up and running.

To test it, I messaged Honey from my desktop: “Can you list out all the skills that you can access right now?” Honey pulled up a list of skills, planning, code quality, iOS, ship, media, everything that was available locally. I swapped to my mobile app, opened the same Honey conversation, and there was the response. Seamless.

Before I dig deeper into Buzz, I want to show you something I switched to recently because it solved a real problem I ran into a few days ago. You might have seen the news: Netnut’s network just got seized by the FBI. Investigators allege it was routing traffic through millions of compromised devices instead of a real opt-in network. If you were relying on Netnut for any AI agents, research bots, or scrapers, you are probably scrambling for an alternative right now. That’s exactly where I was. For a lot of the builds on my channel, I need reliable proxy access, and after this whole mess I actually care where the IPs come from.

I switched to Data Impulse. They run a fully transparent, opt-in residential network: over 90 million IPs in 195 countries, all from real users who knowingly share their bandwidth. No compromised devices, no hidden bundling. The dashboard is straightforward. You create a proxy plan and get your credentials, host, port, username, password. That’s literally everything you need to plug into whatever you’re building. I had a quick script pulling search results through Data Impulse in about two minutes of setup. Request goes out through a residential IP, real data comes back.

What really caught my attention is the pricing. The industry average is $3 to $8 per gigabyte for residential proxies. Data Impulse charges $1 per gigabyte, pure pay-as-you-go, no subscription, and traffic never expires. You literally pay for what you use. If you’re looking for a Netnut replacement, I’ll drop the link in the description. Big thanks to Data Impulse for sponsoring this video.

Alright, back to Buzz.

Managing Projects with AI Agents

Once you start using Buzz, you’ll want to connect your actual codebases. That’s where the project section comes in. You can link Git repositories so that agents know what you’re working on and can help you manage it.

I messaged Bumble and said, “I want to add a repository. Inside my desktop folder there’s an Eric OS project. Create that project in the projects tab and check it out to the main branch.” Bumble kicked off the process, cloned the repo, and announced that Eric OS was live in the Buzz projects. When you open the project view, you see repositories, pull requests, issues, contributions, and all the related message threads. It’s a full development hub.

Then I asked Bumble, “Are you able to see the skills inside that project?” Scrolling back, it had already inspected the top-level folder structure: skills, projects, a full mode folder, even an honest caveat file. It had access to everything.

But here’s where it gets really handy. I told Bumble, “Create a channel just for this project. Name it Eric OS, take the lead, and bring Fizz and Honey into that channel as well.” Bumble did exactly that. It created the Eric OS channel, added the other two agents, and posted an announcement that this channel is now the home for everything related to Eric OS. Now imagine doing this for every project you run: dedicated channels, each with its own set of agents, focused on that single codebase. You can scale that across your whole team.

And speaking of team members, Buzz is not just for AI. You can invite real people too. In settings, there’s an invites section where you copy a link, send it to your teammates, and they jump right in.

Where Everything Lives: Community vs. Agents

A question I had early on was: if I close my computer, can I still use Buzz? The answer is yes for the community side, but there’s a nuance you need to understand.

The community itself, your channels, message histories, memberships, lives on a remote server. So all that persists and you can browse past conversations even from a different device. However, the AI agents (Bumble, Fizz, Honey, and any custom ones you create) are stored on your local computer. If you close your laptop, you can’t talk to them.

The workaround is simple: you can host those agents on a VPS instead of your local machine. Just make sure the VPS has access to Claude Code (or whichever harness you’re using), and configure the agents so they can talk to anyone you intend, not just the server owner. That way your agents stay alive 24/7, and you can reach them from anywhere.

Key Takeaways

Buzz is a group chat platform for teams and AI agents, built by Jack Dorsey (creator of Square).
It uses identity keys instead of passwords: lose your key, you lose access.
You can chat with agents one-on-one or pull several into the same channel for collaborative replies.
The huddle feature lets you jump on a live voice call with any agent, and the full transcript is saved in the thread.
Every agent has persistent memory, so you can teach it personal preferences (like your pizza love) or project context.
Mobile pairing lets you pick up conversations and run agents from your phone.
The projects section connects Git repositories, creates dedicated channels, and lets agents manage code, issues, and pull requests alongside human teammates.
The community lives on a remote server, but agents run locally unless you host them on a VPS for always-on access.

So that’s Buzz. If you’re using Slack or Teams with a sprinkle of AI, this thing might be a serious upgrade. I’d love to know what you think: would you switch over? Leave a comment down below, and I’ll catch you in the next one.