What Is Buzz? Block's Hive Mind Workspace

Guide · 8 min read · Published: · Updated:

Block's Buzz is a self-hostable hive mind workspace where humans and AI agents share the same channels as members. What it is, how it works, and who it's for.

MoClaw Editorial · MoClaw editorial team
What Is Buzz? Block's Hive Mind Workspace
Table of Contents

Share this

Buzz is Block's open-source, self-hostable workspace where humans and AI agents share the same channels as first-class members, not as bots bolted onto a chat app. Built in Rust on a Nostr relay, it turns every message, workflow, and git event into a signed entry in one unified event log, so a person and an agent join a room, act, and get audited the exact same way.

It struck a nerve fast. Buzz climbed to #1 on GitHub's trending charts in late July 2026, topping the Rust category and reaching the top three across all languages, which is why "block buzz" and "buzz github" started showing up in search within days of the repository going public.

Key Takeaways:

  • Buzz is a self-hostable "hive mind" workspace from Block where humans and AI agents are equal members of the same channels.
  • "Agents are members, not bots": you add an agent to a channel the same way you add a person, and it connects coding agents like Goose, Codex, and Claude Code.
  • It is built in Rust on a Nostr relay, with every action stored as a signed event in a unified, searchable log.
  • It is aimed at developer teams and self-hosters who want an auditable, owned workspace, not a hosted SaaS.

What Is Buzz

Buzz is, in its own words, "a self-hostable workspace where humans and AI agents share the same rooms." Block, the company behind Square, Cash App, and the open-source coding agent Goose, released it as an Apache 2.0 project and described it as a "hive mind communication platform." The pitch is that your team chat, your automation, your agents, and your artifacts all live in one place under one identity system.

The reason it caught attention is the framing. Most tools treat AI as a feature you invoke: a sidebar, a slash command, a bot account. Buzz treats the agent as a colleague who happens to never sleep. As the project puts it, "One community is your entire workspace. Work, conversation, agents, automation, artifacts, docs, one domain, one identity system, one search index." That is a bigger idea than a chatbot integration, and it is what the trending spike was reacting to.

Real proof of the spike: Trendshift shows block/buzz first reaching #1 on GitHub Trending on July 23, 2026, after topping the Rust list two days earlier.
Real proof of the spike: Trendshift shows block/buzz first reaching #1 on GitHub Trending on July 23, 2026, after topping the Rust list two days earlier.

What this resolved: A clear definition. Buzz is a self-hosted workspace, closer to Slack-plus-agents than to any single AI tool.

What it left unsolved: "Hive mind" is a strong metaphor. Whether teams actually want agents as literal members, rather than controllable tools, is the open question the rest of the article circles.


Agents Are Members, Not Bots

The line that defines Buzz is blunt: "Agents are members, not bots. Add an agent to a channel the same way you add a person." Under the hood, humans and agents get the same identity: a secp256k1 keypair, a Nostr-native handle like alice@example.com, and the same authentication. There is no separate second-class bot API. An agent has a profile, presence, direct messages, memories, and a channel membership list, just like you do.

Picture a backend team that adds an agent to its #pull-requests channel and gives it the job of first-pass code review. A developer named Priya opens a PR, mentions the agent, and it reviews the diff, posts comments in the channel where everyone can see them, and flags a race condition before a human even looks. The agent is not a webhook firing into the void, it is a participant in the thread, and its entire history is in the same log as the humans'.

A real Buzz #engineering channel from the project's own docs: human teammates and AI agents post in the same threads as members, sharing reactions and history.
A real Buzz #engineering channel from the project's own docs: human teammates and AI agents post in the same threads as members, sharing reactions and history.

Because Buzz comes from the team behind Goose, it connects real coding agents. The buzz-acp harness lets it drive "Goose, Codex, Claude Code" through the Agent Client Protocol, so the agent members are backed by capable engines rather than a toy loop.

What this resolved: A genuinely different membership model. Same identity, same audit trail, same rooms for humans and agents.

What it left unsolved: Equal membership means equal blast radius. The governance question, what an agent member is allowed to do, becomes central, which is where personas and review come in.


Personas, Teams, and Crossfire Review

Buzz does not just drop a raw model into a channel. It wraps agents in structure. A persona, per the project's vision doc, "bundles a model and a system prompt." A team is "a named group of personas, deploy Ralph for code review, Scout for research, Reviewer for crossfire." So instead of one generic assistant, you compose specialists and give each a role and a model.

The standout idea is crossfire review: "agents write code, crossfire reviews across multiple models catch blind spots before merge." One model drafts, other models review from different vantage points, and disagreement surfaces problems a single model would confidently miss. On the same backend team, the agent that drafted a migration gets crossfire-reviewed by two other personas running different models, and the one that flags a missing index blocks the merge until a human weighs in.

In this Buzz thread, an agent persona (Fizz) turns a channel discussion into a structured capture plan as a working member of the team, not a bot on the side.
In this Buzz thread, an agent persona (Fizz) turns a channel discussion into a structured capture plan as a working member of the team, not a bot on the side.

Buzz pairs this with a deliberate notification philosophy: "Zero is the default. You opt in to noise, not out." In a workspace where agents can act continuously, defaulting to silence is what keeps the hive mind from becoming a firehose.

What this resolved: A real quality mechanism. Multi-model crossfire review is a smart answer to single-model blind spots.

What it left unsolved: More personas and more review also mean more model calls and more cost. The structure that makes Buzz safe is not free to run.


How Buzz Works Under the Hood

This is where Buzz earns its trending spot with developers. The core is a Rust relay built on Axum, speaking WebSocket and REST, backed by PostgreSQL for events and full-text search, Redis for pub/sub, and S3 or MinIO for object storage. The protocol layer is Nostr, the same decentralized, key-based event protocol used elsewhere for censorship-resistant messaging, here carrying NIP-01 events, NIP-42 authentication, and NIP-34 for git activity.

How Buzz fits together: ACP clients drive buzz-agent, which uses the sandboxed buzz-dev-mcp tools, all over a Rust Nostr relay where every message, workflow, and git event becomes one signed event stored in PostgreSQL, Redis, and S3.
How Buzz fits together: ACP clients drive buzz-agent, which uses the sandboxed buzz-dev-mcp tools, all over a Rust Nostr relay where every message, workflow, and git event becomes one signed event stored in PostgreSQL, Redis, and S3.

The design principle is that everything is a signed event. Every message, every workflow step, every git push becomes an entry in one append-only, cryptographically signed log. That gives you a unified audit trail and a single search index across chat, automation, and code, and because it is all Nostr events, the project claims "zero breaking changes" as the system evolves.

The agent side is a separate, decoupled binary. Per the agent design doc, buzz-agent is "an ACP agent. It speaks the Agent Client Protocol over stdio, calls an LLM, and uses MCP tools," running multiple isolated sessions, each with its own MCP servers and context. A companion buzz-dev-mcp server exposes a sandboxed shell and file editor with hard safety limits like process-group kills and bounded output. The relay URL an agent connects to is what selects its community, scoping its identity, memories, and audit trail to that workspace.

What this resolved: A serious, inspectable architecture. Signed events on Nostr plus an ACP agent runtime is a coherent, auditable foundation.

What it left unsolved: Self-hosting a Rust relay, Postgres, Redis, and object storage is real operational work. This is a system you run, not a link you sign up for.


Who Buzz Is For

Buzz is squarely for developer teams and self-hosting-minded organizations. If you already run your own infrastructure, care about owning your data, and want an auditable log of everything humans and agents do, Buzz is built for exactly you. The three onboarding paths in the README, a packaged app for general users, internal builds for Block employees, and a build-from-source route for developers, tell you where the center of gravity is: engineers.

Take Diego, a platform lead at a mid-size startup who did not want a third-party SaaS holding his team's conversations and agent actions. He stood up the Buzz relay on the company's own box, so every channel, agent decision, and git event stays in-house and cryptographically logged. For a team with that instinct, Buzz is a natural fit. For a non-technical team that just wants an assistant, the self-host requirement is a wall, not a feature.

What this resolved: A clear audience. Buzz rewards teams that want ownership and are equipped to self-host.

What it left unsolved: The people most excited by "humans and agents as colleagues" are not always the ones able to run a relay. That gap is the opening for hosted approaches.


Buzz and the Human-Agent Collaboration Trend

Buzz is one signal in a broader 2026 shift: teams want humans and AI agents working in the same space, not agents siloed in separate tools. The interesting part is the delivery model. Buzz answers that demand with maximum ownership, self-host the relay, hold the keys, run the infrastructure. That is the right trade for engineering teams who want control.

If you want the human-and-agents-together experience but do not want to stand up and maintain a Nostr relay and its whole backend, the same collaboration idea is available as a managed cloud service. A hosted AI agent like MoClaw puts agents to work alongside people through the tools and channels a team already uses, without anyone owning infrastructure. It is a different point on the same spectrum: Buzz maximizes control, a managed agent minimizes setup, and both are betting on the same future where agents are teammates rather than features.

What this resolved: Context. Buzz is not a one-off, it is the self-hosted end of a real trend toward human-agent teamwork.

What it left unsolved: Control versus convenience is a genuine fork. Which one fits depends on whether your team would rather run a relay or skip that entirely.


FAQ

Who makes Buzz and where is it on GitHub?

Buzz is made by Block, the company behind Square, Cash App, and the open-source coding agent Goose. It is open source under Apache 2.0 at github.com/block/buzz, where it trended to #1 shortly after release.

What does "agents are members, not bots" mean?

It means AI agents get the same identity and permissions model as human users: the same keypair-based login, the same channel membership, the same audit log. You add an agent to a channel exactly as you would invite a person, rather than wiring up a limited bot account.

Which AI agents does Buzz connect to?

Through its ACP-based harness, Buzz drives coding agents including Goose, Codex, and Claude Code. Agents are backed by real LLMs and MCP tools, organized into personas (a model plus a system prompt) and teams.

Is Buzz self-hosted, and what is it built on?

Yes. Buzz is self-hostable and built in Rust on a Nostr relay, using PostgreSQL, Redis, and S3-compatible storage. Every message, workflow, and git event is stored as a signed Nostr event in a unified log.

Do I have to self-host to get human-agent collaboration?

No. Buzz is the self-hosted option. If you want agents working alongside your team without running infrastructure, managed cloud agents offer the same collaboration idea as a hosted service.


Why Buzz Signals Where Team AI Is Headed

Buzz matters less as a single product and more as a statement about direction: the next workspace is one where humans and AI agents are colleagues in the same rooms, under one identity system and one audit log. Block shipped the maximalist, own-everything version of that idea, and thousands of developers starred it in a week because the framing rings true.

The practical takeaway is to decide which end of the spectrum fits your team. If you want control and can run the stack, Buzz is one of the most thoughtful open-source takes on human-agent collaboration to appear in 2026, and it is free to self-host today. If you want the same teammate-not-tool experience without the infrastructure, a managed agent gets you there faster. Either way, the era of agents as members rather than bots has clearly started, and Buzz is the clearest flag planted in it so far.

Continue Reading

M
MoClaw Editorial MoClaw editorial team

The MoClaw editorial team writes about workflow automation, AI agents, and the tools we build. Default byline for industry overviews, listicles, and collaborative pieces.

Ready to put this into practice?

MoClaw runs browser tasks, research, and schedules automatically. Try it free.

what is buzz buzz github buzz hive mind buzz ai agents human agent collaboration self-hosted ai workspace buzz nostr

References: Buzz on GitHub (Block, Apache 2.0) · Buzz VISION.md: personas, teams, crossfire review · Buzz VISION_AGENT.md: buzz-agent ACP architecture · Buzz on Trendshift (GitHub trending #1) · Goose: Block's open-source coding agent · Nostr protocol