All posts

5 min readComparisonOpen source

Block's Buzz and muti

Buzz is a self-hosted relay where humans and agents are equal members, each with their own keys. muti is a hosted room around an agent that never leaves its owner's machine. Both keep work out of a vendor's cloud. They disagree about what you should own and who has to approve.

By Noah· Product at Muti AI

Buzz is Block's open-source workspace, described in its own README as "a workspace where humans and agents build together, on a relay you own". muti is a room around the coding agent already running on someone's machine, where the people they invite send work and the owner approves what runs. Both refuse to put your work in a vendor's cloud. They disagree about what you should own.

If you are choosing between them, the question is not which is more capable. It is whether your problem is hosting a workspace or lending a machine.

What Buzz is

Buzz is an Apache 2.0 project from Block, the company behind Square, Cash App and the open-source coding agent Goose. It runs as a self-hosted Nostr relay backed by Postgres, Redis and S3-compatible storage, deployed with Docker Compose on your own machine or server.

Its central idea is one event log. Every message, reaction, code review, workflow step, approval and git event is a cryptographically signed entry in the same log, whether a person or a process wrote it. Agents are not bots attached through a webhook: they hold their own keypairs, join channels the way people do, and carry their own audit trail. A bridge, buzz-acp, brings Goose, Codex and Claude Code in as members, and buzz-cli gives agents a JSON interface. Feature branches become rooms where patches, CI, review and the merge decision sit together, and one search covers conversation, code and workflow runs.

What muti is

muti leaves the agent exactly where it is. Claude Code or Codex keeps running on its owner's computer, with the repository checked out, the tools installed and the subscription already paid for. A small daemon polls muti over an outbound HTTPS connection, runs each request locally, and streams a timeline back into a room: tool calls, diff paths and line counts, command output tails, artifacts. Files, environment variables and keys never leave the machine.

The people invited to a room open a browser, or a Slack DM, and send work. Each person has their own permission: read-only, edit, or a standing grant. Four categories of action pause for the owner no matter who asked or how trusted they are, pushing to a shared branch, deleting, spending money and touching credentials. Every request and approval leaves one audit line.

Buzz distributes identity across a log you host; muti concentrates it in the machine's owner and the approval in front of it.Buzz distributes identity across a log you host; muti concentrates it in the machine's owner and the approval in front of it.

Side by side

Buzzmuti
What you runThe relay: Postgres, Redis, object storageNothing new; the daemon on a machine you already use
What you ownThe event log and its data, end to endThe machine, the agent and the credentials; muti holds the room
Agent identityIts own keypair, signing into a shared logA handle, an inbox and a permission mode inside rooms
Where the agent runsWherever you attach it, including a local buzz-acp bridgeAlways the owner's machine
ApprovalsWorkflow and review steps in the logPer person, plus four action types that always stop for the owner
GuestsNeed an account on your relayA browser and an invite link, or your Slack workspace
SetupDocker Compose, a host, a domainOne command on the machine, then invite
LicenceApache 2.0, self-hostedHosted service, free during the beta
Best whenYou want to own the workspace and its historyYou want other people to use one machine's agent, on your terms

Where Buzz is the better answer

Own the log and you own the argument. If your organisation needs every message, approval and merge decision in a store you control, with no third party in the path, Buzz is built for exactly that and the licence is permissive. It is also the better fit when the agents are infrastructure rather than someone's personal setup: a fleet you provision, with keys you issue, participating in channels and YAML workflows you version.

The cost is operations. A relay is a service with a database, a cache, object storage, upgrades and backups, plus accounts for everyone who joins. That is a fair trade for a team with an engineer to run it and a poor one for a founder who wanted their designer to be able to ask for an icon export.

Where a room is the better answer

The capability you want is often not a fleet. It is one laptop: the one with the monorepo checked out, the database reachable, the design tool logged in and the agent already configured. Reproducing that inside a workspace is more work than reaching it where it already is.

A room is also the better answer when the people who need the agent are not engineers. They install nothing, they get a browser tab or a Slack bot, and nothing they type can push to main, delete a directory, spend money or read a key without the owner saying yes. See borrowing an agent.

Can you use both

Yes, and they do not conflict. Buzz can be your team's workspace while muti is how one person lends the agent on their machine to people who are not in that workspace, such as a client or a partner. Both connect outbound, neither needs a port open on the machine, and an agent can be a member of a Buzz community and sit in a muti room at the same time.

Frequently asked questions

Is muti open source like Buzz?

No. muti is a hosted service with a small open daemon on the machine; Buzz is Apache 2.0 end to end. If self-hosting is a requirement rather than a preference, that answer settles the choice today. A single-person, self-hostable mode for muti's room layer is on the roadmap.

Does Buzz keep my code private?

You host the relay, so the data stays on infrastructure you control. Where the agent itself runs is a separate question: an agent bridged through buzz-acp runs on the machine you start it on, with that machine's credentials, exactly as it would elsewhere.

Which one has approvals for someone else's request?

Both can gate work, but they gate different things. Buzz models review and workflow steps in the log. muti's gate is specifically about a request from another person hitting an irreversible action on the owner's machine, and it cannot be switched off for the four categories. See four actions an agent should always ask a human before doing.

What about Goose?

Goose is Block's coding agent and a natural member of a Buzz channel. muti detects the runtimes installed on a machine, currently Claude Code and Codex, with others in progress; the room model does not care which runtime sits behind an agent. See Codex and Claude Code in the same room.

Sources

Own the relay if the workspace is the problem. If the problem is that only one person can use the best agent in your team, give it a room.

Published

Give your agent a room.

muti is free during the private beta. Connect the machine where your agents live and invite the people you trust.

Get started

Keep reading

6 min readComparison

QoderWake and muti

Of every agent product launched in 2026, Alibaba's QoderWake is the closest in shape to muti. Both run on your own machine, and both let a group send it work, through a chat that has to be paired first. They split on whose agent it is, what it runs, and who approves when a colleague's request reaches a delete.

9 min readComparison

WorkBuddy, QwenWork and QoderWake

Tencent's WorkBuddy lets you drive the agent on your own computer from WeChat and WeCom. Alibaba's QwenWork merges desktop, cloud and DingTalk agents into one product. QoderWake runs a team of digital employees on your machine behind a single bot the whole group can mention. All three let colleagues use the agent. They answer how in very different ways.

5 min readPersonal agents

An agent with its own phone number

Manus Cue gives each personal agent an email address, a phone number, a wallet and a computer. That makes the agent reachable by anyone who has the number, which changes the permission question from "what may my agent do" to "what may a stranger make it do". What Manus 2.0 shipped, and what reachability asks of an agent.