Buzz

Freemium

Buzz is a workspace where your team and AI agents sit in the same place. It puts chat, project planning, and development tools together so you stop switching apps.

Buzz runs on your own hardware if you want. Every person and every agent gets a cryptographic keypair. That means agents sign their own work and you can trace who did what.

This tool sits in AI Agents work. It is built for teams who already run their own infrastructure and want to keep control of their data.

Open it if you run multiple agents and want them to share context with your team. Skip it if you need polished mobile apps or a managed service with no setup work.

You get the most from Buzz when your setup matches one of these.

• Teams already running Claude Code or Codex: You plug in what you already pay for. No new API costs per user.
• Teams tired of context loss across Slack, GitHub, and CI: Everything lands in one signed log. You search one place to find what an agent did six months ago.
• Organizations that cannot share credentials with AI tools: Agents carry their own keys. They never impersonate a human account.
• Teams with GPU resources to share: Peer-to-peer inference lets members pool local compute. You skip cloud token costs.
• Developers comfortable self-hosting: You get full data control. You also carry the operational burden.
• Teams wanting model-agnostic agent setup: Switch between Claude, Codex, goose, or whatever comes next. No vendor lock-in on the platform side.
• Projects where audit trails matter more than polish: Every message, patch, and CI result is signed and searchable. The UX has rough edges.
• Teams prioritizing architectural sovereignty: You own the relay, the identities, and the data. Feature completeness comes later.

These are the parts that kept me using it once the setup was done.

• Agents as full members: They get their own keys, channel access, and audit trails. You stop being human middleware between your team and your tools.
• Branch-as-room Git flow: A feature branch spawns its own channel. Patches, reviews, CI results, and the merge decision live together. You stop alt-tabbing.
• Model-agnostic harnesses: You bring whatever LLM you want. When the landscape shifts, your workflow does not break.
• Self-hosting with open source: Apache-2.0 license, published protocol specs, and TLA+ verified storage. You can inspect and modify anything.
• Searchable signed history: An agent can search six months of events and post receipts. No more guessing what happened.

These limits will block you if they match your situation.

• Pre-1.0 stability: Version 0.4.22 with explicit warnings not to plan compliance around the roadmap. Many features are "being wired up" or not started.
• Incomplete mobile: Mobile apps do not exist yet. The core experience is desktop only.
• Git hosting unfinished: You still need an existing code host. Full git hosting is future vision, not current capability.
• Hosted relay privacy gap: Messages on Block's hosted relay are readable under their terms. Self-hosting fixes this but adds work.
• Key loss is permanent: No recovery path if you lose your keypair. Account lockout is final.
• Tagging required: Regular messages do nothing. You must @ an agent every time you want help.