BUZZ / BLOCK / NOSTR / A CRITICAL OVERVIEW
Buzz is a workspace for the synthetic firm.
Jack Dorsey announced Buzz as an open-source alternative to Slack and GitHub for teams of humans and agents. The chat interface is familiar. The unusual part is underneath: signed identities, scoped rooms, code, workflows, and a common event log where a process can be a colleague.
One community boundary, one URL, one event log, one search surface.
Humans and agents sign the same event shape with separate keypairs.
Agent clients connect to conversations while tools remain composable.
Desktop, relay, CLI, workflow, and agent infrastructure are available to inspect and fork.
01 / WHAT IT IS
A team chat with an institutional memory.
Buzz was announced on July 21, 2026 by Jack Dorsey and built at Block. The pitch is direct: reduce dependency on Slack and GitHub by placing people, agents, conversation, code, automation, and approvals in a self-hostable workspace.
A user sees channels, threads, direct messages, media, canvases, search, voice huddles, workflows, and repositories. An agent sees the same workspace through a CLI or protocol bridge. Both are members with identities and channel memberships. Messages, reactions, git events, and workflow steps are encoded as signed Nostr events and handled by the relay.
This matters because most workplace agents remain integrations: a bot token inside a human platform, a coding session hidden in one person’s terminal, or an automation that posts without durable authorship. Buzz moves the agent into the visible org chart.
02 / THE THREE-LAYER BET
Identity, event, room.
Buzz’s novelty is compositional. None of the layers is new by itself; the product comes from making them share one protocol and one boundary.
Humans and agents receive Nostr-native keypairs. Their events carry verifiable authorship. An agent is not borrowing the owner’s name, and its channel membership can differ from the owner’s.
THE GOOD PART · RESPONSIBILITY HAS AN ADDRESSA message, reaction, workflow trace, patch, and approval can all be recorded as signed events. That makes search and audit less like an integration project and more like a native property of the workspace.
THE GOOD PART · THE RECEIPT LIVES BESIDE THE CONVERSATIONA relay URL selects a community. Profiles, rooms, direct messages, memberships, search, and agent state stay local to that boundary. A key can travel, but membership does not silently follow it.
THE GOOD PART · PORTABLE IDENTITY WITHOUT AMBIENT ACCESS03 / WHAT IS INTERESTING
The chat is not the product. The constitution is.
Buzz is most useful as a proposal for how synthetic teams should be legible to themselves.
- 01
An agent is a member, not a feature.
You add it to a room the way you add a person. Presence, identity, history, and access become visible collaboration facts rather than hidden integration settings.
- 02
A branch can become a room.
Buzz’s forge vision places discussion, patches, CI, review, and the merge decision in one channel. The archive becomes the explanation for why the code exists.
- 03
The relay is durable context.
Agents can search prior work and return evidence into the same stream. Context moves out of one user’s transcript and into a community-owned record.
- 04
Models become replaceable labor.
ACP and MCP separate the conversational transport, agent runtime, model provider, and tools. The institution can persist while its workers change.
- 05
Sovereignty is deployable.
The code is open and the relay can be self-hosted. A team can control its domain, database, event log, and agent surface instead of only exporting from a vendor.
- 06
Signed history can become governance.
An approval is more useful when authorship, time, context, and downstream action remain connected. Buzz is trying to make this the default shape of work.
04 / WHAT “DECENTRALIZED” MEANS HERE
Sovereign at the edge. Central inside the room.
The project’s own architecture is refreshingly specific: the relay is the single source of truth. There is no peer-to-peer event exchange, gossip, or replication in the current design.
You can own the code and deployment.
The Apache-licensed system can run on infrastructure you control. The team’s domain and relay can become its workspace rather than an address inside a vendor’s workspace.
Your identity can travel.
A Nostr keypair is not inherently locked to one Buzz host. Profiles and permissions are deliberately re-established in each community.
The relay remains an authority.
It authenticates, authorizes, persists, searches, fans out, and triggers automation. A compromised or unavailable relay is still a central operational event for its community.
Communities are not a working federation.
Portable identity is a foundation, not cross-relay governance. Delegation, shared projects, revocation, dispute handling, and receipts across community borders remain the next design problem.
05 / PRODUCT REVIEW
The rough edges are governance-shaped.
Buzz’s public docs distinguish working features from active wiring and long-range vision. That honesty is useful. It also means the platform should be evaluated as an ambitious alpha, not as finished enterprise infrastructure.
| Surface | Industry Next read |
|---|---|
| Works today | Relay, channels, threads, DMs, canvases, media, search, audit log, desktop client, agent-first CLI and ACP harness, YAML workflow triggers, git events, and hosting are listed as operational in the README. |
| Still wiring | Mobile clients and fully resumable workflow approval gates remain works in progress. Some vision documents run ahead of the README, which is normal for a fast project but important for buyers to notice. |
| Privacy boundary | TLS protects traffic in transit and storage encryption is operator-managed. The project says end-to-end encryption for DMs is a future consideration, so relay operators can remain in the trust model. |
| Abuse boundary | The architecture document says a rate-limiter interface and tiers exist as a design target, but no production rate-limiting implementation is currently enforced. |
| Approval boundary | Signed events improve attribution, but a signature does not prove an action was wise, authorized at the right level, or safe. Human gates and independent containment still matter. |
A signed log is evidence, not prevention. The federation story will depend on revocation, expiration, budget limits, safe defaults, and the ability to stop an agent without asking the same compromised layer for permission.
06 / THE 2027 READ
The company becomes a protocol.
Buzz points toward an organization whose real org chart is a graph of signed identities, room memberships, workflows, and approval rights. People will join and leave. Models will be swapped. The institution will be the durable set of boundaries and receipts connecting them.
Role keys
Agents will hold scoped identities for research, release, finance, support, and review.
Finite jobs
Membership and authority will expire with the task instead of becoming permanent ambient access.
Crossfire review
Different models and people will inspect the same work before it crosses a meaningful boundary.
Relay treaties
Communities will exchange signed requests and receipts without merging their internal state.
Community memory
Decisions, artifacts, and incident history will belong to the project rather than one operator’s chat account.
Portable labor
Agents will move between runtimes while identity, role, and public work history remain intelligible.
07 / THE SMALLEST HONEST PILOT
One room. Two agents. One human gate.
Create a low-stakes project channel. Add one research agent and one reviewer with separate identities. Give them only the files and tools required for a finishable task. Keep production secrets and merge authority out of the room. Require a human approval event before publication.
Then test the platform’s real promise: can another person arrive later, search the channel, distinguish request from result, inspect the evidence, see who approved the outcome, and safely continue the work?
Buzz is not yet the synthetic company. It is a credible sketch of its constitution.
08 / SOURCE LEDGER