I have a pile of model subscriptions, and what I actually wanted was simpler than another chat window: a desktop room where those agents could talk to each other while I brainstorm, and where the coding, the research and the rest of the work could happen in the same place. I asked Claude. It first pointed me at group-chat apps and at Goose, Block’s own agent harness. When I named Buzz, that was the closer match for the original ask.
Buzz is Block’s shared workspace for people and AI agents. Jack Dorsey launched it in July 2026 as a free, open-source group chat for teams of people and agents, built to cut a team’s dependence on stitching Slack and GitHub together by hand. You download it at buzz.xyz. The current desktop app, Buzz 1.0, is out for macOS, Windows and Linux, with mobile apps as well. The project’s README still says 1.0 is in beta. The release notes, platforms, plugins, Huddles, Attention watches and hosted pricing are covered in the 1.0.0 write-up.
we’re launching BUZZ!
July 21, 2026
a new groupchat platform for teams of people and agents of all sizes, built to reduce our dependency on slack and github. model-agnostic, decentralized, self-sovereign, and open source. 🐝
What it is
Think of a familiar team chat, then give every agent a seat in the channel. Block’s launch post describes channels, threads, direct messages, voice, media, code repositories and automated workflows, built on the Nostr protocol. Each participant, human or agent, holds a cryptographic keypair that belongs to them, not to the app they happen to be using. Agents can post, review code, run approved automations and talk alongside people, rather than living in a private side window you have to ferry notes in and out of.
The workspace is model-agnostic and agent-agnostic. Official docs for the 1.0 desktop app list local harnesses for Claude and Codex (on macOS or Linux, with the provider’s own sign-in), Goose (included with Buzz, including on Windows), and Pi. Anything that speaks the Agent Client Protocol can live in the room. You can also run a hosted agent where your account allows it. Skills, MCP servers and tools stay with the harness the agent is already running; Buzz is the room, not a second copy of every integration.
You can use Block’s hosted relay or run your own. The code is Apache-2.0. The 1.0 desktop work lives in the block/buzz-app repository; the older 0.X line remains in block/buzz and is still supported.
Why it’s handy
The useful idea is almost rude in its simplicity: stop being the copy-paste middleman.
Tyler Longwell, writing on Block’s engineering blog, describes the pattern most of us already know. People work alone with an agent in a little harness window, copy the output into a team chat, then paste the replies back into the harness. Agents make individual work faster and teamwork slower, because every handoff needs a human to ferry context between tools. In Buzz, he writes, the people, the agents, the decisions and the work share one room.
He often runs one stronger agent as a lead, with cheaper, faster agents researching, building, testing and reviewing in parallel. They talk through ordinary mentions in the channel. Humans can redirect the work while it is happening, and the visible history keeps the failed paths as well as the final call. A feature or a bug can get a short-lived channel; discussion, patches, CI, review and the signed merge decision share one record. Search later and you find why the obvious fix was wrong, not only the green tick.
Agents also have their own identities. They do not borrow your face or your login. The agent’s owner signs a narrowly scoped authorisation, and the agent signs its own work. If a key leaks, you revoke the agent without replacing the human identity behind it.
That is the utility I was looking for: a place where agents can talk amongst themselves, and where I can still be in the conversation.
What people are doing with it
Block is using the product to build the product. The official Buzz account posted a short look at design in the same conversation: Juno shares a first pass, the team gives feedback, the next version comes back into the same thread. At AGNTCon and MCPCon, the team also showed engineering, design and agents shipping together in one shared space.
Building Buzz inside Buzz.
October 28, 2026
Juno shares a first pass. The team gives feedback. The next version comes back into the same conversation.
A short look at how we design together.
In another official clip, Ken Lopez, Wes Billman and an agent called Forge work together to build an app in Buzz. The point of the demo is the unglamorous one: the agent posts directly in the channel, so nobody has to sit in the middle copying messages from a private session into the team chat.
Whose agent is this?
October 29, 2026
Watch @kenlopez15, @wesbillman and Forge work together to build an app in Buzz
Agents speed up the work. Buzz speeds up the teamwork. Agents posts directly in channels, so you stop being the copy-paste middleman.
Thomas Petersen, a principal designer at Block who works on Buzz, posted an early example of the same network effect: he wanted to build a feature, and his agent found that someone else on the network had already built and shipped it. Plugins stretch that further. The 1.0 app can take custom pages, panels and tools, including things like maps in a conversation, and a wider marketplace is still on the way. Official support docs are blunt about the risk: plugins are trusted code, not sandboxed extensions, so you install only what you trust and inspect the source when you can.
Outside Block, the most complete integration I’ve seen is from Nous Research. In late July it said Hermes Agent now runs Buzz, with three paths: Buzz Desktop can auto-discover a local Hermes install, a relay bridge can host an agent identity, and a native gateway can treat Buzz as a first-class messaging platform alongside the usual chat apps. Official Hermes docs match that split, and recommend the native gateway when you want Hermes’ own memory, skills, approvals and sessions to stay intact.
Hermes Agent now runs Buzz.
July 29, 2026
The self-hostable workspace from @blocks puts humans and agents in the same messaging channels and codebase.
Three ways to use Buzz with Hermes (and vice versa):
– Buzz Desktop auto-discovers your Hermes install runs it locally
– A relay bridge gives…
On Hacker News, Longwell described the everyday loop in plainer terms still: spin up a channel, add your team and your agents, send them out to research, then implement, test and review in the same room, screenshots and all. That is also the loop I wanted when I went looking.
How to try it
Download the app from buzz.xyz, create or join a community, and add two agents before you add twenty. Put them in one channel with you, give them distinct jobs, and let them talk in the thread. If you already use Claude or Codex on a Mac or a Linux machine, sign in through Buzz’s own integration. If you want a Windows-friendly local agent, start with Goose. If Hermes is already your operator, follow Nous’s gateway docs rather than assuming the desktop auto-discover path will keep all of its features.
Keep the computer and the app running if you want local agents, including Attention watches, to stay awake. Use a self-hosted relay if the content is sensitive; official docs note that messages in hosted communities are not end-to-end encrypted. Don’t leave a flock of agents mentioning each other with no limits, and don’t install a plugin you haven’t looked at.
Buzz is young, and the README still calls 1.0 a beta. The room is the useful part. That is what I went looking for, and that is what it is built to be.
Discover more from Sulat AI & Tech News
Subscribe to get the latest posts sent to your email.
