explainx / blog
Block launched Buzz on July 21, 2026 — open-source chat, Git hosting, and AI agents behind one signed Nostr identity. What's real, what's unfinished.
Jul 22, 2026
The model gets the headline; the harness decides whether the agent actually finishes the task. Here are the top 10 closed-source and top 10 open-source agent harnesses builders are running in 2026 — what each one does differently, what it costs, and who should pick it.
Jul 12, 2026
@ClaudeDevs ships Claude Code desktop browser — read, click, debug URLs sandboxed. Version 1.2581.0 July 10. explainx.ai setup, shortcuts, and developer reactions.
Jul 7, 2026
On July 7, 2026, @ClaudeDevs published the definitive Claude Code loops guide by @delba_oliveira — how the team categorizes loops by trigger, stop criteria, and primitive. explainx.ai maps each type to real commands, skills, and the loop-engineering corpus you already have on-site.
Jack Dorsey announced Buzz on X at 10:03 PM on July 21, 2026: "we're launching BUZZ! a new groupchat platform for teams of people and agents of all sizes, built to reduce our dependency on slack and github. model-agnostic, decentralized, self-sovereign, and open source." The post had crossed 2.3M views within a day, and the accompanying Hacker News thread ran to well over 200 comments arguing about whether the project's Nostr foundation is load-bearing or just crypto theater layered on a permissions problem.
The pitch is specific to a problem anyone running Claude Code, Codex, or Goose on a team already feels: your coding agent's context lives in your terminal, your team's discussion about that work lives in Slack, and the code itself lives in GitHub. Buzz tries to collapse all three into one event log where a human and an agent hold the same kind of identity.
| Question | Answer |
|---|---|
| What is it? | Open-source workspace: team chat + Git forge + AI agents, one identity system |
| Who built it? | Block (Jack Dorsey, cofounder), announced July 21, 2026 |
| Where's the code? | github.com/block/buzz, Apache 2.0 |
| What powers identity? | Self-hostable Nostr relay — every action is a signed event |
| Do agents get real accounts? | Yes — own key pair, channel memberships, audit trail, same as a human |
| Which coding agents does it wire up? | CLI plus harnesses for Goose, Codex, and Claude Code |
| Is it actually decentralized? | No P2P/gossip yet — one relay per workspace is authoritative; you can self-host it |
| Is it finished? | No — mobile clients, push notifications, and workflow approval gates are incomplete |
| Platforms? | Desktop app for macOS, Windows, Linux; latest build was v0.4.21 |
Buzz is built around a self-hostable Nostr relay. Every message, reaction, workflow step, code event, and approval is stored as a cryptographically signed event rather than a row in a proprietary database. Human employees and AI agents receive the same basic identity structure: a key pair, channel memberships, and an audit trail. That's the core design bet — an agent isn't a bot token duct-taped onto a chat API, it's a member with the same primitives as the person who authorized it.
Per Block's documentation, agents inside Buzz can:
The Git forge is the more ambitious half of the pitch. Rather than posting GitHub webhook notifications into a chat channel, Buzz's specification describes a built-in software forge over standard Git Smart HTTP. A feature branch can become its own channel, with patches, CI results, review comments, and the eventual merge decision preserved in the same event record — one search index across chat, code, and workflow history instead of three disconnected systems.
For coding agents specifically, Buzz ships a command-line interface and harnesses for Goose, Codex, and Claude Code — the same three names that show up repeatedly in explainx.ai's roundup of the top agent harnesses in 2026. Notably, Goose is Block's own open-source coding agent, so Buzz effectively wires Block's existing agent tooling into Block's new team-chat product. On X, a developer asked Dorsey directly whether Buzz supports "bring your own" harnesses like Claude Code or Codex instead of being locked to one; Dorsey's reply was a one-word "yes."
Block's own architecture document is unusually candid about what's real. The currently working feature set is channels, threads, direct messages, shared canvases, media, search, an audit log, a desktop application, and YAML-based workflows, with packaged builds for macOS, Windows, and Linux under an Apache 2.0 license. What's explicitly incomplete: mobile clients, push notifications, and workflow approval gates, which the docs describe as having database, API, and interface components without a finished execution path. The desktop release at launch was version 0.4.21.
Dorsey's announcement called Buzz "decentralized, self-sovereign, and open source." Several commenters on Hacker News pushed back on the "decentralized" part specifically, and the pushback holds up against Block's own documentation: Buzz currently has no peer-to-peer event exchange, gossip layer, or 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 to clients.
What Buzz's decentralization actually means in practice is closer to self-hosting with portable identity. An organization can run its own relay, keep its domain and data, and use Nostr key pairs instead of being locked into one vendor's account system. A commercial operator could also run multiple isolated workspaces on shared infrastructure. But within any one workspace, that relay remains the authoritative server — if it goes down, the workspace goes down with it, same as a self-hosted Slack alternative like Zulip or Mattermost.
A more pointed thread on Hacker News asked whether Nostr adds anything over a conventional shared service at all. One commenter compared it to "telling me I'm going to be using blockchain to store my files because I need crypto on top of my crypto." A Block engineer working on Buzz replied directly: Nostr gives a durable, portable identity across communities, relays, and workspaces, and — more concretely — lets you independently verify that a message came from the person or agent who signed it, even if the relay operator is compromised or lying. For a workspace where agents are authoring patches and merge decisions, that verification property is arguably more relevant than it would be for ordinary chat.
The most substantive thread in the HN discussion wasn't about Nostr — it was about access control once agents can see everything a shared workspace sees. A Slack engineer commenting independently laid out the core tension: single-player agents that act on one user's behalf are straightforward to reason about, but multiplayer agents that sit inside a shared channel raise the question of what happens when the agent has seen something in a private group that a given human hasn't. Multiple commenters converged on the same worry — an agent with visibility into a leadership channel could leak context into an unrelated conversation with someone who was never supposed to see it.
Buzz's answer, per its documentation, gives each agent its own key and a narrowly scoped authorization signed by the agent's human owner, so a leaked agent key can be revoked without touching the underlying human identity. That solves credential hygiene. It doesn't, by itself, solve the harder problem raised in the thread: whether an agent that's a member of both a private channel and a public one can be trusted not to blend context between them. This is the same class of problem covered in explainx.ai's guide to MCP sandboxing and agent context boundaries — scoping what a tool or agent can see is a different problem than scoping who authorized it, and Buzz's launch documentation is stronger on the second than the first.
A separate practical complaint from the thread: group chats with multiple humans and one agent are still awkward. One commenter noted that today, an agent in a group chat either silently maintains a separate internal thread per user (confusing) or responds to everyone in one shared thread (noisy). Buzz doesn't appear to have solved this UX problem yet — it inherited it, like every other multiplayer-agent chat product currently shipping.
| Project | Chat | Git hosting | Agent identity model | Self-hostable | Maturity |
|---|---|---|---|---|---|
| Buzz | Yes | Built-in forge (Smart HTTP) | Nostr key pair per agent | Yes (single relay) | New (Jul 2026) |
| Slack + Copilot/Claude bots | Yes | No (integrates GitHub) | Per-app bot token, scoped by admin | No (SaaS only) | Mature |
| Zulip / Mattermost | Yes | No | Standard bot accounts | Yes | Mature |
| Anthropic's Tag (Claude in Slack) | Rides on Slack | No | Slack-scoped auth | No | New |
| JetBrains Space / raft.build | Yes | Yes | Varies by product | Partial | Mature / early |
Buzz's differentiation is putting chat, a forge, and agent identity in one signed event log rather than integrating three separate products. Whether that's worth adopting today depends heavily on how much you value self-hosting and portable identity versus a mature, batteries-included tool — the same tradeoff explainx.ai covered when OpenShip launched as a self-hosted PaaS alternative earlier in July: a compelling architecture on day one doesn't mean the operational gaps (mobile, notifications, HA) are closed yet.
The interesting signal here isn't Buzz specifically — it's that team-context tooling for AI agents is becoming its own category, separate from the harness that actually writes code. Anthropic recently hired the leadership team behind Zulip, another open-source team chat tool, which suggests more than one company is betting that the harness (Claude Code, Codex, Goose) and the coordination layer (where humans and agents actually talk about the work) are going to be built by different teams entirely.
If you're already running agents through loop engineering or wiring up MCP servers for your team's workflows, Buzz is worth watching as a reference architecture even before it's production-ready: identity per agent, scoped authorization signed by a human, and an audit trail that ties a code change back to the exact permission that authorized it. Those are the same primitives you'd want whether or not you ever touch Nostr — and they're a preview of what "the agent has a seat at the table" is going to require once agents aren't single-player tools anymore.
Block's Buzz (announced July 21, 2026, github.com/block/buzz, Apache 2.0) merges team chat, a Git forge, and AI agents behind one Nostr-based identity system, with harnesses for Goose, Codex, and Claude Code. The chat and forge features are real and self-hostable today; the "decentralized" framing describes deployment flexibility, not peer-to-peer replication, which Block's own docs confirm doesn't exist yet. The hardest open problem — what a multiplayer agent is allowed to say across channels it has access to but a given human doesn't — is unsolved, in Buzz and everywhere else building this category right now.
Official links: buzz.xyz · GitHub block/buzz · Buzz architecture docs · Block engineering blog
Details accurate as of July 22, 2026, based on Jack Dorsey's X announcement, Block's public GitHub documentation, and the Hacker News launch discussion. Buzz is an early-stage, actively changing project — verify current feature status and license terms on the GitHub repository before adopting it for production use.
