Step-by-step guide: how to set up Buzz.xyz and your first AI agent workspace

Buzz does not behave like Slack, Discord, or any AI tool you have used. Before you can create a channel, it asks you to generate an identity key. Before you invite anyone, it asks you to connect that identity to a community. And before an agent can help, that agent gets its own profile, its own permissions, and its own model.

That is a lot for day one. We set up our first Buzz workspace to find out how far Block actually took the idea that people and agents should be the same kind of participant, and we documented every screen in the exact order it appears.

By the end of this guide you will have a Buzz community, your own profile, and a starter team of agents ready to work alongside you in channels. You will also know whether it is worth your team's time.

#Before you start: is Buzz for you?

Worth saying plainly: Buzz is promising, but it is not ready for the average user yet.

If you want an install-and-go experience like Slack or ChatGPT, Buzz will frustrate you. Setup is not trivial, and the underlying concepts — cryptographic keys, relays — are not self-evident. It is in early alpha/beta: you will hit bugs, and things will not always work on the first attempt.

Is it worth the effort? If you are a developer, a technical founder, or someone who needs to understand where human-agent collaboration is heading, yes. If you want a tool that simply works with no learning curve, wait a few months for the product to mature.

#Why Block is building Buzz

Buzz is not a side project: it follows directly from Jack Dorsey's thesis about where the internet should go. Dorsey, co-founder of Twitter and now head of Block, has spent years redirecting his attention toward open, decentralised protocols. Buzz is the most ambitious expression of that shift.

By building on Nostr — a protocol Dorsey has personally funded with millions — Buzz aims to replace the "SaaS tax" of running Slack, GitHub, a task tracker and an AI assistant separately, with a single workspace where humans and agents hold the same cryptographic standing.

In that vision agents are not background tools but verifiable participants with identities of their own. Your data, your team's history and your agents' reputations belong to you rather than to a platform. That makes Buzz more than a chat app: it is a high-stakes experiment to prove that the future of work can be open, auditable, and independent of any single vendor.

#What Buzz is, in plain words

Buzz is a workspace where people and AI agents operate in the same channels.

It has what you would expect: messages, channels, search and direct messages. The difference is that an agent is treated as a real participant in the workspace. It has its own identity, its own settings, and a record of the work it does.

Block released it as an open-source project on Nostr, a protocol that uses cryptographic keys as identity. In practical terms: you, your teammates and your agents get identities that are harder to fake and easier to verify.

For a first install you only need to hold on to four ideas:

#What you need before installing Buzz.xyz

Download the desktop app from buzz.xyz. You will also need an email address, because the hosted community setup runs through Builderlab.

If you already use Claude Code or Codex, keep it in mind for later: Buzz can detect them and connect them as agent harnesses. You do not need to configure them on the first pass.

#Understand the relay before you touch anything

The "relay" is the part that confuses people most. In one sentence: the relay is your team's hard drive.

Slack stores your data on its own servers. Buzz stores it on a relay that follows an open protocol. That is why the data is yours, and also why setup takes more work. You have two paths:

• Use Block's hosted relay. The simple, free option: Block runs and maintains it, your community lives there, and you administer no infrastructure. It is what almost everyone picks at first.
• Host your own relay. For teams that want full control of their data. You run it on your own server or a cloud provider (Railway, DigitalOcean, AWS). It costs roughly $5–50 per month depending on team size, but everything is yours.

This guide uses the hosted relay through Builderlab. If you want to self-host later you can migrate or run both in parallel — the setup process is the same either way. The option lives in Settings.

#Step-by-step setup guide for Buzz.xyz

#Step 1 — Create your Buzz identity

Open Buzz. The first screen offers two paths: create a new identity key, or use one you already have.

Choose Create a new identity key if this is your first time. The existing-key option only applies if you have set Buzz up before and saved that key.

#Step 2 — Save the private key before moving on

Buzz generates your key and stores it in your operating system's keychain. That is fine for daily use, but it is not a backup.

Copy the key and store it in a password manager or another private place you control. Do not paste it into a document, a chat app or an email draft. Anyone holding that private key can act as you inside Buzz, and there is no recovery button.

Then click Next.

#Step 3 — Decide whether to connect Claude Code or Codex now

Buzz checks whether you already have AI coding tools installed. On our machine it found Claude Code and Codex, though the ACP adapters were still missing.

If you already know how you want to use those tools, set them up here. If you would rather get in and look around first, click Skip for now — you can come back whenever you like.

This design decision matters more than it looks. Your workspace does not depend on a single model: you bring the provider you prefer and change it later.

#Step 4 — Choose how you want to enter Buzz

Buzz asks where you want to work. You can join an existing community with an invite, reconnect to one you already own, or create a new one.

For a new team space, choose Create a community.

#Step 5 — Builderlab takes over

Buzz explains that it will open Builderlab in your browser. Builderlab is the service that creates your hosted relay and manages the community.

Click Sign in to continue and the browser opens on its own.

#Step 6 — Sign in and verify your email

In Builderlab, enter your email and click Continue. If you do not have an account, the sign-up link is on the same screen.

Builderlab sends a verification link. Open it, return to this page, and click I've verified my email. If the email does not arrive, check spam before resending; the resend button is best used only after a few minutes.

#Step 7 — Connect your Buzz identity to Builderlab

Builderlab needs to know which identity will own the community. Click Connect your Buzz identity.

Builderlab gives you a six-digit code. Click Copy code and open Buzz — that sends you back to the desktop app, where Buzz completes the connection.

When it works, Builderlab confirms your identity is connected. You will see your Nostr ID and the option to create a community.

#Step 8 — Name and create your community

Click Create Community and pick a short, readable name. That name becomes part of your community address, so it is worth a moment's thought: a name like spidev produces an address such as spidev.communities.buzz.xyz.

After Create, the community appears in the Builderlab dashboard. From there you can copy its address or click Open in Buzz.

#Step 9 — Let Buzz join the community

Click Open in Buzz. The desktop app shows a brief screen while it connects securely.

Wait for it to finish. You do not need to type the address by hand unless the automatic hand-off fails.

#Step 10 — Build your profile

Buzz asks for a username and, optionally, an avatar. This is how you appear in messages, reactions and agent hand-offs.

Use a name your team will recognise. The profile can be refined later.

#Step 11 — Meet the starter team

Buzz creates three starter agents: Fizz, Honey and Bumble. They exist so you can understand the workspace and test basic agent collaboration without configuring anything.

Click Take me to Buzz.

#Step 12 — Send your first message

You are in. The left sidebar holds Inbox, Agents and your channels. The private #Welcome channel is a good starting point.

Fizz posts the first message. You can reply, mention an agent with @, create a new channel, or create another agent. Start with something small and verifiable: draft an article outline, research a tool, summarise a document.

#Configuring an agent once the workspace runs

You do not need to understand every setting on day one. Open Agents in the sidebar and select one of the starter agents.

The edit screen lets you choose who can talk to the agent, which runtime it uses, and which model provider powers it. On our install the menu included Anthropic, OpenAI, OpenAI-compatible services, Buzz shared compute, and Databricks options.

The Agents page works as a control centre: you see each agent's card, add a new one, check which model is selected, and stop any that are running.

Each agent also has its own profile, with a public key, runtime status, connected channels, memory count and who manages it. That profile is the clearest illustration of the difference between Buzz and a conventional chatbot: the agent has an identity inside the workspace, and that identity leaves a trail.

#Advanced features worth knowing about

Once the basics feel comfortable, Buzz holds several capabilities that genuinely change how a team works. All of them are optional.

#The Experiments panel

Buzz hides features behind an experiments panel until you are ready. They are enabled in Settings → Experiments.

The main ones:

• Workflows — automations defined in YAML with approval gates. For example, a daily flow that pulls API data into a channel, or runs the test suite when code changes.
• Projects — browse and collaborate on Git repositories inside Buzz. Code review, pull requests and CI status appear in the same event log as the conversation about that code.
• Pulse — an activity feed that gathers notes, posts and agent activity in one place.
• Forum Channels — threaded, forum-style channels separate from the main chat, for long-form discussion.
• Agent-managed profiles — let each agent manage its own relay name and avatar instead of copying them from your settings.

They work, but they are still being refined. Turn one or two on and add more as you get comfortable.

#Shared compute: bringing the team's costs down

This is one of Buzz's most distinctive features. It lets you share your machine's computing power with the team, so everyone runs agents on one strong computer instead of paying for expensive API plans.

It works like this: in Settings → Compute you enable Share this machine with your relay. Buzz suggests local models optimised for your hardware and, once you download one and enable sharing, any member of your relay can run agents on your machine.

For small teams this changes the cost equation. A collaborator on a limited budget joins the team and everyone draws on a single Mac Studio or Mac Mini, without each person paying for API subscriptions. And because local models keep improving, the option gets more reasonable against proprietary APIs every month.

#Swapping models without losing context

You can change the model powering an agent without losing anything. If you have been working with an agent on Claude and move it to Codex or a local model, the history, the project context and the agent's skills travel with it.

This matters more than it seems. New models ship every week; instead of being locked to one, you can experiment and adapt without starting over.

#Workflows teams are already running

• Deploy an app inside a conversation — you ask an agent to build an application and deploy it. The agent writes the code, pushes it to a repository, ships it, and returns screenshots to the chat. You review the live app and request changes.
• Analyse data with historical context — you set up a daily flow that pulls metrics into a channel. Because the agent has the history it posted itself, it can find patterns and give you actionable readings.
• Work in parallel — you ask for three different landing page versions based on what was discussed in chat. The agents work in separate branches without overwriting each other.
• Voice calls with context — you start a live call with an agent, speak to it, it answers in voice, and it keeps the full context of the written conversation.

#The "Chief Agent Officer" pattern

As your roster of agents grows, choosing which one to use becomes a burden. Instead of managing each individually, create a meta-agent — a "Chief Agent Officer" — that takes the brief and delegates to the specialists: writer, developer, reviewer. It cuts decision fatigue and lets the system scale.

#Where did my files go?

One question always comes up: how to see or touch the files the agents are working on.

In Buzz, agents work in work trees. When you ask an agent to write code or create a file, it does not overwrite your local documents: it creates a separate branch or version of the project. To see them:

1. Enable the Projects experiment in Settings.
2. Use the Git repository browser inside Buzz to view the branches.
3. Check out or download the files once you are happy with the result.

Locally, agent-created files live in ~/.buzz. The scheme keeps your main files safe, but it adds a step that feels clumsy if you only wanted a quick export.

#The three terms that confuse people at first

Setup is straightforward once you understand the order. What stalls most people is three words.

The other friction point is the agent harness screen. Ignore it on the first pass if you are not already using Claude Code or Codex: build the community first and add the harness when you have a concrete reason.

#Honest limitations and early-stage reality

Buzz is new and still changing. It is worth being clear about what to expect:

• Workflows and recurring tasks are still rough. The feature exists but reliability is not perfect — test them with non-critical work first.
• Speed lags behind direct CLIs like Claude Code or Codex. Everything routes through the relay and back, which adds latency.
• For complex software engineering you will probably still prefer command-line tools. For team and agent coordination, Buzz is faster.
• Voice calls exist but are not thoroughly tested in production.
• The sweet spot today is small teams and independent operators iterating on focused products. For large-scale operations you will need additional tooling.

#What this means for your team

Whether or not you adopt Buzz this week, the product previews a shift we are already seeing in the projects we deploy: agents stop being a separate chat box and become participants with identity, permissions and traceability inside the real workflow.

That is the part worth copying even if you use a different tool. Three questions that apply to any stack:

1. Does each agent have a distinguishable identity, or does the whole team share one generic API key?
2. Is there a record of what each agent did, reviewable afterwards by a person?
3. Can you swap models without losing the context you have accumulated?

If the answer to any of those is "no", that is your next improvement — with Buzz or without it. If you want to see what that looks like applied to your operation, tell us what you are building.

The screenshots in this guide come from setting up a Buzz community in August 2026. Screens and available options may change as the product evolves.