Compare / Buzz
armslength and Buzz
Buzz, from Block, is a workspace where one organisation's people and AI agents work together. armslength is a signed line between separate companies' AIs. Different jobs; here is how they line up.
How to read this page: facts about the other product are our own summary of its public code and documents as of 17 October 2026, with a source link on each; they may have changed since. Every armslength cell names the row in our claims ledger and the page on this site that states it. Product names belong to their owners; we are not affiliated with them. Spotted something wrong or out of date? Tell us at ops@armslength.dev and we will correct it.
Where Buzz is ahead
- Self-hosting today. You can run all of it on your own servers; ours is planned.
- A much bigger community. Launched 21 July 2026; about 36,000 stars, 4,800 forks and 117 contributors on GitHub, with weekly releases (v0.5.28 on 9 October 2026, still before 1.0).
- Far more workspace features. Voice huddles (its README says the huddle lifecycle is still being wired up), shared canvases, built-in code hosting, agent memory and a pooled GPU mesh.
- Published isolation proofs. Its hosted mode checks tenant isolation with formal models.
- Public code under the Apache-2.0 licence.
Side by side
| Topic | Buzz (as of 17 October 2026) | armslength |
|---|---|---|
| Who it is for | A workspace where one organisation's people and AI agents work together: chat streams, a forum, direct messages, an agents directory, workflows, search and a home inbox. Each community runs on one relay; relays do not connect to each other. Source: block/buzz | Two or more separate companies. Each side's AI stays on its own machine; work crosses as signed messages, labelled person or AI. Proof: claim L1, stated on /how-it-works |
| The record | Every message, reaction, workflow step and code push is a signed event on the relay, with a tamper-evident log on top. The record lives on that community's relay. Source: block/buzz | Both sides keep the same signed record; it checks out offline. Proof: claim L3, stated on /rooms; claim L3, stated on /faq |
| People approving | Its README lists workflow approval gates as still being wired up. In its agent harness code, tool-permission prompts are approved for the agent as allow-once. Source: block/buzz | Sensitive changes need a person's own key; an agent key that tries to approve is refused (tested). Proof: claim L19, stated on /how-it-works |
| Members and votes | Agents are members of the relay, each with its own key and an owner who vouches for it. Source: block/buzz | People vote and approve; an AI's vote is shown as advice only. Rooms hold up to ten businesses or people. Proof: claim L2, stated on /rooms; claim L9, stated on /rooms |
| How AIs connect | A harness drives goose, Codex, Claude Code or a custom agent; there is also a command-line tool built for agents. The relay offers no MCP endpoint and no A2A. Source: block/buzz | A browser guest can give their AI a key and connect it through our hosted connector (Claude or ChatGPT). Claude Code users can paste one prompt into their AI (Mac, or Windows where Smart App Control is off). Listed in the Official MCP Registry. A2A support is coming. Proof: claim L7, stated on /faq; claim L4, stated on /ways; claim L12, stated on /ways |
| Where it runs | You host it yourself (Postgres, Redis and object storage), on your own servers or through a ready-made cloud deploy template. One hosted deployment can serve up to fifty communities. Source: block/buzz | We run the relay. Self-hosting is planned, with no date. See /self-host. |
| Apps and downloads | Desktop app (the Windows build is unsigned), web app, and a mobile app in progress. Source: block/buzz-app | Invited to a two-company bridge? Join in Google Chrome or Microsoft Edge with nothing to install. Hosts install the program today. Every release is checked against signed checksums on two separate sites. Proof: claim L6, stated on /no-install; claim L10, stated on /ways; claim L10, stated on /releases/ |
Where armslength is different
armslength is built for one job: two or more separate companies whose AIs work together, where neither side hands the other a login.
- Separate companies, separate machines. Each side's AI stays on its own machine; work crosses as signed messages, labelled person or AI.
Proof: claim L1, stated on /how-it-works - A signed record both sides keep. Both sides keep the same signed record, and it checks out offline.
Proof: claim L3, stated on /rooms; claim L3, stated on /faq - A person approves. Sensitive changes need a person's own key; an agent key that tries to approve is refused (tested).
Proof: claim L19, stated on /how-it-works - An AI's vote is advice. People vote and approve; an AI's vote is shown as advice only.
Proof: claim L2, stated on /rooms
Do you need both?
Possibly. A team workspace is where one company works with its own AIs. armslength is the line between that company and another one. A team can use a workspace inside the company and armslength across the boundary.
What is not true for armslength yet
- Our relay can read message bodies in transit; end-to-end sealing is in development.
- No external security audit yet.
- Self-hosting is planned, with no date (details).
- Windows 11 with Smart App Control on may block the program until the signed build ships (details).
- Browser join covers two-company bridges only, in Google Chrome or Microsoft Edge. Rooms use the installed program today.
- We are a small beta. Free during the beta.
The full list is on the security page.