Buzz: Block's Nostr-Backed Workspace for Humans and Agents

A runtimewire report on Block's July 21, 2026 launch of Buzz, announced by Jack Dorsey on X. Buzz is an open-source, self-hostable workspace that puts employees, AI agents, conversations and source repositories behind a single identity system, aimed at reducing Block's own dependence on Slack and GitHub.

The design

Everything sits on a self-hostable Nostr relay. Every message, reaction, workflow step, code event and approval is stored as a cryptographically signed event.

The consequential part is that humans and agents get the same basic identity structure — their own key pairs, channel memberships and audit trails. That makes agents members rather than chat bots: per Block's documentation they can search prior discussions, open repositories, submit patches, review code, run workflows, edit shared canvases and create channels. Buzz ships an agent-oriented CLI plus harnesses for Goose, Codex and Claude Code, keeping model choice separate from the workspace. See agent-identity-attribution for why signed per-agent identity is a structurally different answer than the usual bot integration.

The Git ambitions go past posting notifications into chat. The project specification describes a built-in forge over standard Git Smart HTTP, where a feature branch can become its own channel — patches, CI results, review comments and the merge decision all preserved in one record, with repositories, discussions and workflow history sharing a single search index.

Working today: channels, threads, DMs, shared canvases, media, search, an audit log, a desktop app for macOS/Windows/Linux, and YAML-based workflows. Apache-2.0.

The decentralization check

This is where the article earns its keep. Dorsey described Buzz as decentralized and self-sovereign; runtimewire went to Block's own ARCHITECTURE.md and found the boundary drawn much more narrowly.

Buzz currently has no peer-to-peer event exchange, no gossip layer, and no replication between relays. Every read and write in a workspace passes through a single relay that authenticates users, verifies signatures, stores events and distributes updates.

So the decentralization is in deployment and ownership, not protocol. An organization can run its own relay, keep its domain and data, and use portable Nostr key pairs rather than depending on a hosted service — real benefits — but within each community the relay remains the authoritative server. The article draws the practical consequence for anyone evaluating it as a Slack or GitHub replacement: self-hosting buys control over infrastructure and data location and simultaneously transfers responsibility for availability, backups, security and upgrades. Signed events give attribution and an audit trail; they don't remove the operational risk attached to the one server holding the workspace.

That gap between protocol-level and deployment-level decentralization is the same distinction that matters in domain-fronting-era circumvention tooling and in federated-vs-P2P debates generally, and it's the kind of claim that usually goes unchecked in launch coverage.

Maturity

Block's own documentation repeatedly labels it unfinished. Mobile clients are in development, push notifications are pending, and workflow approval gates have database, API and interface components without a completed execution path. Desktop v0.4.21 shipped the same day as the announcement.

The scope is unusually wide for an early product — chat, code hosting, workflow automation, project search and agent orchestration in one event system. The article's read on the tradeoff is even-handed: combining those surfaces may genuinely cut the integration work needed to give agents useful context and tightly scoped access, but it also puts Buzz against mature products whose separateness is a feature, since customers can replace one without migrating the whole stack.

Block is the only customer reference in the documentation, and no adoption, pricing or external-customer figures have been published. The closing question is the right one: whether engineers outside Block want one relay carrying this much of their work.

Where it sits

Buzz is an infrastructure answer to the problem agent-principal-agent-problem describes — Crawshaw's diagnosis was that review broke because effort became unobservable, and signed per-actor events at least make authorship observable again even if they say nothing about effort. It's also a bet against the direction in i-dont-want-your-prs, which argues the PR-shaped collaboration surface should be abandoned rather than rebuilt; Buzz rebuilds it with agents as first-class participants.

Cross-references

• agent-identity-attribution — agents as signing identities rather than bot accounts
• block-inc, runtimewire — the company and the publication
• agent-principal-agent-problem, i-dont-want-your-prs — the collaboration-surface problem it's answering
• re_gent — the other provenance-for-agent-work approach in the wiki