Buzz is Block's self-hostable team workspace built on the premise that AI agents should be members of a room rather than bots attached to one. Released in March 2026, it has climbed to roughly 13,400 stars with 1,106 forks and gained over 12,500 stars in the past week alone, making it one of the fastest-moving Rust projects on GitHub right now. The README is unusually candid about the crowded category it enters — "Yes, it's another AI-adjacent developer tool. We're sorry" — and then makes a specific architectural argument for why it is different. ## One Event Log, Two Kinds of Author The core design decision is that Buzz is a Nostr relay. Every message, reaction, workflow step, review approval, and git event is a signed event in a single log. The consequence is that a human and an agent are structurally identical to the system: same event shape, same identity model, same audit trail, different keypair. That is not a cosmetic distinction. Agents get their own keys, their own channel memberships, and their own audit trail, which means access is scoped by identity the way a teammate's is, rather than by a permission flag bolted onto a bot integration. You can let an agent triage a bug without handing it blanket workspace access, and the record of what it touched is the same record that covers what people touched. A Buzz **community** is the workspace reached by URL, and in the shipping single-relay setup the relay URL selects exactly one community. Hosted operators can serve many communities across domains while keeping that semantic boundary — tenant-observable state stays community-local even when Postgres, Redis, and object storage are shared underneath. ## What Agents Can Actually Do The claim that separates Buzz from chat tools with an LLM sidebar is surface area. Inside a Buzz community an agent can open repos, send patches, review code, run workflows, edit canvases, orchestrate other agents, join voice huddles, create channels, and pull in whoever needs to see the result. Git integration runs on NIP-34, so patches, repo announcements, and CI status arrive as events in the same log as the conversation about them — a feature branch becomes a room where the patch, the review, and the merge decision share a searchable history. The agent-facing entry point is `buzz-cli`, deliberately built as JSON in, JSON out for LLM tool calls, plus `buzz-acp`, an ACP harness that bridges to Goose, Codex, and Claude Code. Automation runs through YAML workflows triggered by messages, reactions, schedules, or webhooks. The stated end state is a release that writes itself: a workflow fires on a tag, an agent drafts notes from merged PRs, a human reacts with a thumbs-up, and it ships — every step signed and searchable. ## Architecture and Honesty About Status Buzz is a Rust workspace of focused crates around a single source of truth, the relay. `buzz-core` holds zero-I/O types and Schnorr verification, `buzz-relay` is an Axum WebSocket and REST server, and services split into `buzz-db` on Postgres, `buzz-auth` for NIP-42/98, `buzz-pubsub` on Redis, `buzz-search` on Postgres full-text search, and `buzz-audit` maintaining a hash-chain log. Media goes to S3 or MinIO via Blossom. The desktop client is Tauri plus React. The README carries a three-column table splitting features into what works today, what is being wired up, and what is opinion without code yet. Relay, channels, threads, DMs, canvases, media, search, audit log, the desktop app, `buzz-cli`, YAML workflows, and git hosting are in the first column. Mobile clients, workflow approval gates, and huddle lifecycle events are in the second. Web-of-trust reputation and push notifications are in the third, with a note asking readers not to plan a compliance program around it. That kind of disclosure is rarer than it should be. ## Pros and Cons The strengths start with the identity model — giving agents real keys and real audit trails is a more coherent answer to agent permissions than most tools offer. Collapsing chat, git, CI, workflows, and search into one event log removes the glue code teams normally write between those systems. Apache 2.0, fully self-hostable with a production Compose bundle, and 50 contributors with releases landing roughly weekly. The caveats deserve weight. Nostr as a substrate is an unusual bet, and it commits you to an ecosystem most teams have no experience operating. The self-host path needs Docker, Rust 1.88+, Node 24+, pnpm, and Hermit, which is heavier than a container pull. 710 open issues against a four-month-old project reflects both velocity and rough edges. Mobile clients do not exist yet. Windows requires Git for Windows so the agent shell tool can resolve bash. And Block maintains a separate internal build, so the OSS release is not the version its largest user runs — a divergence worth watching. ## Outlook The category Buzz is entering — agent-native team workspaces — is where much of the tooling effort is going in 2026, and most entrants are chat apps with an agent bolted on. Buzz's differentiation is that the substrate was chosen first and the workspace built on top, which is harder to retrofit than to start with. Whether Nostr proves to be the right substrate is the open question; whether the identity-first framing of agent access is correct seems considerably more settled. ## Conclusion Buzz is worth evaluating for teams already running agents against their codebase who are frustrated that those agents live outside the record of the work. It suits organizations that want self-hosted sovereignty over that record. Teams that need mobile clients, a light operational footprint, or a settled API should watch it rather than adopt it this quarter.