Jack Dorsey’s Buzz Takes On Slack and GitHub: Here’s What It Actually Does

Jack Dorsey just launched a free, open source rival to Slack and GitHub called Buzz, built for teams where humans and AI agents work in the same channels. Here is what it actually does, how it compares to Slack’s own AI push, and how to try it.

Jack Dorsey wants out of Slack and GitHub, and he is not being subtle about it. On Tuesday, his payments company Block launched Buzz, a free workspace where employees and AI agents share the same channels, the same threads, and the same audit trail.

Dorsey’s pitch on X was blunt: model agnostic, decentralized, self sovereign, open source. Translation: no vendor lock in, no single company holding every key, and agents that behave like teammates instead of chatbots bolted onto the side of a chat window.

Here is what Block actually shipped, how the agent side of it works, how it stacks up against Slack’s own AI overhaul, and how to start poking at it yourself if you run a small team that is tired of gluing five tools together.

What Block Actually Launched

Block, the company also behind Square, Cash App, Afterpay, and Tidal, released Buzz as a free and open source workspace under an Apache 2.0 license. The short version, pulled from Block’s own GitHub repository and its launch coverage:

One workspace, several tools merged. Buzz bundles channels, threads, direct messages, voice huddles, canvases, media sharing, search, and an audit log, plus early Git hosting so pull requests and code review can happen in the same window as the conversation about them.
Agents join as members, not bots. Every human and every agent gets its own cryptographic identity with defined permissions. Agents can post messages, review and submit code, trigger workflows, and even open their own channels.
It runs on any model. Teams can plug in agents built on Anthropic’s Claude Code, OpenAI’s Codex, or Block’s own open source agent framework, goose, first released in January 2025.
It is genuinely early. Desktop apps are available now for macOS, Windows, and Linux. Mobile clients, push notifications, and full Git integration are still being built.
You can self host it or let Block run it. Teams choose between a fully self hosted deployment or Block’s managed version, with the source code public either way.

Dorsey framed the whole thing as a way to cut Block’s own reliance on Slack and GitHub. In the company’s launch post, Block’s head of AI capabilities, Bradley Axen, said “every company is going to need a place where humans and agents work together.” His point was that the real fight from here is over who controls that place, a Salesforce style vendor or an open protocol nobody owns outright.

How the Agent Side Actually Works

The interesting engineering choice here is not the chat interface. It is the identity layer underneath it.

Buzz is built on Nostr, a relay based protocol where every participant, human or agent, holds a keypair that belongs to them rather than to the platform. Every message, reaction, workflow step, code review, and Git event gets recorded as one signed event in a single log, whether the author is a person or a process. That gives you one hash chained audit trail instead of a pile of separate logs across five different tools.

In practice, that means an agent working inside Buzz does not get a blanket bot token that can read everything in a channel. It gets scoped permissions and its own activity history, the same way you would scope a new hire. Teams set channel level or thread level rules for what each agent can see and do, and every action that agent takes is traceable back to its own identity rather than a shared service account.

Agents are not limited to answering questions either. According to Block’s own account of the build, agents inside Buzz can open repositories, submit and review patches, run approved automations, join voice huddles, and spin up their own channels when a task needs one. Lucas Isaza, who works on AI and product at Block, has argued that tools like Asana, Jira, and Linear ended up shaping how teams work as much as the work itself, because the tool baked in a specific process like sprints or kanban boards. Strip that out, he says, and the constraint shifts from the tool to the operating model your team chooses.

Buzz vs Slack vs Centaur

Dorsey is not the only person building an AI native alternative to Slack right now. Paradigm’s partner and CTO, Georgios Konstantopoulos, released a similar open source project called Centaur back in May, which he describes as a self hosted virtual employee that runs either inside Slack or through an API. And Slack itself has not stood still. Salesforce rolled out a major AI overhaul to Slack in March, turning Slackbot into a desktop agent that can transcribe meetings, act as an MCP client, and route work to Agentforce agents across the Salesforce ecosystem, running on Anthropic’s Claude under the hood.

Here is how the three actually compare on the things that matter if you are choosing between them:

Factor	Buzz	Slack + Agentforce	Centaur
Who owns it	Open source, Apache 2.0, self host or Block managed	Salesforce, proprietary, cloud only	Open source, self hosted
Agent model support	Any model or harness, including Claude Code, Codex, goose	Built around Agentforce, plus third party agents via MCP	Any model, described as a self hosted virtual employee
Code hosting	Built in, still early	None native, connects out to GitHub or similar	None native
Identity and audit model	Cryptographic identity per human and agent, single signed event log	Standard enterprise permissions and admin controls	Depends on self hosted setup
Maturity	Early stage, launched July 2026	Mature product with a large existing user base	Early stage, launched May 2026
Cost	Free, self hosting costs your own infrastructure	Paid, tiered by plan	Free, self hosting costs your own infrastructure

The honest read: Slack is not short on AI agents anymore, it is short on being open about how those agents are built and who controls them. Buzz and Centaur are both betting that startups building agent heavy workflows will care more about that openness than they will about Slack’s fifteen year head start on polish and integrations.

How to Start Using Buzz

If you run a small team, especially one already leaning on AI agents for coding or ops work, here is the fastest way to see what Buzz actually feels like.

Decide hosted or self hosted first. Block’s managed version gets you running in minutes with no infrastructure to maintain. Self hosting gives you full control over your data and relay, but it means running your own Postgres, Redis, and search stack, so it suits a team with some DevOps comfort already.
Download the desktop app. Grab the build for macOS, Windows, or Linux from buzz.xyz or the releases page on GitHub.
Install and launch it. The desktop app opens straight into workspace setup on first run.
Create your workspace identity. Buzz generates a keypair for you rather than a username and password. That keypair is your identity across the workspace, so treat it the way you would treat any private key.
Invite your team. Add the people who will actually use this daily before you add any agents, so you have a real channel structure to test against.
Connect an agent. Set the BUZZ_PRIVATE_KEY environment variable and use buzz-cli, the agent facing command line tool that speaks JSON in and JSON out, to bring in an agent running on Claude Code, Codex, or goose.
Set permissions per channel or thread. Decide what each agent can see and do before you let it loose, the same way you would scope access for a new contractor.
Give it one real task before you judge it. Point an agent at an actual bug report or a real feature branch rather than a toy example. The audit log and the branch as room pattern are the parts worth stress testing, not the chat UI.

The Honest Limitations

Buzz is not something you should move a live team onto this week, and Block does not pretend otherwise.

Mobile is not ready. Only desktop apps exist right now. If your team lives on phones between meetings, you will feel that gap immediately.
Git hosting is still early. The pitch of replacing GitHub outright is closer to a roadmap item than a finished feature today.
Decentralized has a caveat. Each workspace currently runs on a single relay, which the project’s own documentation calls its single source of truth. That is a more centralized setup than the marketing language suggests, at least for now.
The security question is real and unresolved. An agent sitting in a channel can read everything posted there. Buzz’s answer is scoped identities and full audit trails rather than blanket bot tokens, but whether that holds up under real multi tenant use at scale has not been tested in public yet.
Self hosting is not for everyone. Running your own relay, database, and search stack is a real operational commitment. Teams without in house DevOps capacity will likely lean on Block’s managed version instead, which reintroduces some of the dependency this project is supposedly trying to remove.

The Bottom Line

Buzz is a genuinely interesting bet, not a finished product. The idea that a startup’s chat history, code review, and agent activity should live in one signed, searchable log instead of five disconnected tools is a real insight, and it is one that plenty of founders will recognize from their own tool sprawl.

But Slack is not the sleepy incumbent this launch makes it sound like. It has spent the last year turning Slackbot into a genuine agent platform with deep Salesforce and MCP integration, and most teams already have years of institutional history sitting inside it. Buzz’s real audience right now is smaller and more specific: AI native startups with no legacy Slack workspace to migrate away from, engineering teams comfortable running their own infrastructure, and anyone who wants their agents to carry a portable, verifiable identity instead of a bot token that only means something inside one company’s servers.

Worth installing on a side project this weekend. Not worth betting your whole team’s workflow on yet.

If your team has ever lost a decision because it was scattered across Slack, a GitHub PR, and three DMs, this is the first tool in a while actually trying to fix that instead of adding a fourth place to look.

Sources and further reading

TechCrunch, Jack Dorsey is taking on Slack with Buzz, a group chat platform for teams and their AI agents (July 21, 2026).
Decrypt, Jack Dorsey’s Block Launches Buzz, a Nostr Based Slack and GitHub Rival for AI Agents (July 22, 2026).
Block, Introducing Buzz: Where Humans and Agents Work Together.
GitHub, block/buzz repository and README.
Northeast Times, Jack Dorsey launches Buzz, an open source rival to Slack and GitHub, on the Hacker News reaction (July 22, 2026).
Paradigm, Open sourcing Centaur: multiplayer, self hosted, secure agents.
TechCrunch, Salesforce announces an AI heavy makeover for Slack, with 30 new features (March 31, 2026).