# Block's Buzz puts AI agents on Nostr with their own keys — here's what actually ships
> Block's Buzz makes AI agents full Nostr members with their own keys — but mobile and approval gates aren't shipped yet.
**Source:** Broke It (https://brokeit.dev/posts/block-s-buzz-puts-ai-agents-on-nostr-with-their-own-keys-here-s-what-actually-sh/)  
**Published:** 2026-07-23
**Original source:** https://techcrunch.com/2026/07/21/jack-dorsey-is-taking-on-slack-with-buzz-a-group-chat-platform-for-teams-and-their-ai-agents/
**Topics:** ai-agents, open-source, collaboration, decentralized, developer-tools

---

Buzz is real code you can run today on macOS/Windows/Linux, but half the pitch is roadmap. What ships: chat, a Git forge, workflow automation, and AI agents as full members with their own Nostr keypairs. What doesn't ship yet: mobile, and — this is the big one — approval gates. So you get autonomous agents with cryptographic identity before you get the guardrails to constrain them.

The interesting design choice is that agents aren't bots you `@mention` to run a command. Each agent holds its own Nostr keypair, so it's a peer in the room, signing its own messages. That's a genuinely different model from Slack's app integrations or GitHub's Actions bot that posts as `github-actions[bot]`.

## What actually ships today?

Desktop apps are out. The platform merges group chat with a Git forge and automated workflows, and it wires up harnesses — Claude Code, OpenAI's Codex, and Block's own Goose — so an agent can pick up work and push commits. It's Apache-2.0, which matters: you can fork it, self-host it, and audit what the agents are actually signing. According to <a href="https://techcrunch.com/2026/07/21/jack-dorsey-is-taking-on-slack-with-buzz-a-group-chat-platform-for-teams-and-their-ai-agents/" rel="nofollow noopener">Block's launch coverage</a>, it's pitched as cutting your dependence on Slack and GitHub, which is a big claim for a v1 desktop-only release.

## What broke / what's missing?

Approval gates aren't done. Read that again in the context of "agents are full members with their own keys." A first-class member that can commit code and act on Nostr, with no shipped mechanism to gate its actions, is a trust problem you're building yourself until the gates land. In practice that means today you're either running agents in a sandbox you control, or you're trusting the harness config and hoping. Mobile is also absent, so the collaboration story is desk-only for now.

The Nostr foundation is a double-edged thing. Decentralized identity means no central admin to revoke a compromised agent key the way you'd disable a Slack app. Key rotation and revocation become your responsibility. That's the tradeoff for not being locked into one vendor's identity plane.

## Should you use it?

If you want to experiment with agents-as-peers and you can self-host and sandbox, yes — clone it and poke at the Git forge integration. If you're looking to replace Slack and GitHub for a real team, no, not yet: no mobile, no approval gates, and the security model leans on you. I'd run it in a throwaway workspace with agents that can't touch anything I care about, watch how the Nostr identity model behaves under key rotation, and revisit when approval gates ship. The idea is sharper than most agent-collab demos; the missing guardrails are exactly the part that's hard to build.

## FAQ

### Is Buzz free and open source?
Yes. Buzz is released under Apache-2.0, so you can self-host, fork, and audit it. Desktop apps for macOS, Windows, and Linux are available now.

### How are AI agents different from Slack bots in Buzz?
Each agent gets its own Nostr cryptographic keypair and acts as a full member signing its own messages, rather than a command-triggered integration posting under an app identity.

### Which agent harnesses does Buzz support?
Buzz supports Claude Code, OpenAI's Codex, and Block's own Goose, letting an agent pick up work and push commits through the built-in Git forge.

### What's not ready yet?
Mobile apps and approval gates are still in development. The missing approval gates are the notable gap, since agents already have autonomous, key-backed identities without a shipped way to constrain their actions.
