Type: GitHub Repository
Original Link: https://github.com/block/buzz
Publication Date: 2026-08-18
Summary #
Introduction #
Imagine being in a team meeting where you’re discussing a critical bug. Someone says “Hey, let’s ask the agent to search through our tickets from the last six months to find similar patterns”. The agent not only responds, but opens the repository, reads the code, proposes a patch, submits it for review, and documents everything in the same channel where you’re chatting. It’s not science fiction: it’s what Buzz does every day. It’s a workspace platform that transforms how teams and AI agents collaborate, building together on infrastructure that you control completely.
Buzz is not a simple chatbot integrated into an IDE, nor is it another SaaS platform where your data lives on someone else’s servers. It’s something more radical: a self-hostable workspace where humans and AI agents literally share the same room, with the same identity, the same audit trail, and the same access to repositories. Every message, every reaction, every code approval is an event signed in an immutable log. It’s built on Rust and uses Nostr as its foundation, which means security and transparency aren’t afterthoughts, but part of the architecture itself.
What It Does #
Buzz is essentially a collaborative workspace where the boundary between chat, project management, and software development disappears. When you add an agent to a channel, you’re not adding a bot that just reads: you’re adding a team member with their own cryptographic keys, granular permissions, and complete audit trail. The agent can read the project history, search for specific information with “receipts” (meaning links to threads where it found the information), open repositories, analyze patches, and even execute workflows.
The magic lies in unification. Traditionally, when you develop software, conversation lives in Slack, code in GitHub, workflows in CI/CD, and decisions are scattered everywhere. In Buzz, all of this happens in the same place. When you create a feature branch, you can transform it into a “room” where the patch, CI results, reviews, and merge decision live together as a single coherent story. Looking for something? You search for the conversation, the patch, the workflow run, and the approval simultaneously, because they’re all the same type of event in the underlying log.
Why It’s Amazing #
The ‘wow’ factor is that agents aren’t guests, they’re citizens. They don’t have limited permissions or artificial sandboxes. They have the same surface-level capabilities as humans: they can create channels, manage canvases, execute workflows, participate in huddles. But everything is tracked. Every agent action is signed with its key, so you know exactly who did what and when. It’s not “the bot messed up”, it’s “agent X did Y at 14:32 UTC, here’s the cryptographic signature”.
Reasoning with historical context: When you ask an agent a question, you don’t get a generic answer. The agent searches six months of project history, finds relevant threads, and answers you with “receipts” — meaning links to threads where it found the information. It’s like having a colleague who not only knows the answer, but also knows where you first discussed it and shows you. This eliminates the classic LLM problem: hallucinations. If the agent doesn’t find something in the project history, it says so clearly.
Triaging without giving away the keys to the kingdom: One of the thorniest problems with AI agents is security. If you give an agent total access to the repository, it can cause damage. Buzz solves this with identity and granular permissions. You can give an agent permission to read tickets, analyze bugs, and propose patches, but not permission to merge directly. The agent has its own audit channel, its specific permissions, its cryptographic identity. It’s like having an intern who can do research and propose solutions, but you make the final decisions.
Immutability as a feature, not a constraint: Why does Buzz use Nostr (a signed event log protocol)? Because every technical decision becomes a permanent record. When someone approves a patch, that approval is signed. When an agent suggests a solution, it’s tracked. If in six months you wonder “why did we make this choice?”, the answer is in the log. It’s not a secondary feature, it’s the foundation of the platform.
How to Try It #
The code is on GitHub (github.com/block/buzz) and is open source under Apache 2.0. If you want to try it locally, setup is simplified thanks to a Justfile that automates most of the work. You need Docker (for services), Rust (for compilation), and Node.js (for the desktop app). Once you’ve cloned the repository, just run just setup to configure everything: Docker, database migrations, desktop app dependencies. Then just relay starts the relay (the central server), and just dev launches the desktop app.
The documentation is rich: there’s a VISION.md explaining the general philosophy, VISION_SOVEREIGN.md for the self-hosting model, VISION_PROJECTS.md for how projects work, VISION_AGENT.md for agent behavior, and ARCHITECTURE.md for those who want to understand how it’s built internally. CONTRIBUTING.md is well-structured if you want to contribute. It’s not a one-click demo, but the setup is designed for developers who know what they’re doing. If you have questions, the repository has 2810 open issues, which means the community is active and developers respond.
Final Thoughts #
Buzz represents a paradigm shift in how we think about remote teams and automation. It’s not “let’s add a chatbot to our workflow”, it’s “let’s completely rethink the workflow around a space where humans and agents reason together”. With 28,000 stars on GitHub, it’s clear the community sees the potential. Of course, there are legitimate concerns about UX (some describe it as visual “horror”) and privacy when agents have access to sensitive conversations. But these are implementation problems, not conceptual ones.
What makes Buzz interesting is that it’s not another SaaS platform promising to solve everything. It’s infrastructure that you control, built on principles of immutability and transparency, where AI agents aren’t black boxes but traceable citizens of your workspace. If you’re tired of jumping between Slack, GitHub, Jira, and Discord, and want a space where conversation, code, and decisions live together, Buzz deserves serious consideration.
Use Cases #
- Private AI Stack: Integration into proprietary pipelines
- Client Solutions: Implementation for client projects
- Development Acceleration: Reducing project time-to-market
Third-Party Feedback #
Community feedback: The community appreciates the concept of integrating chat, AI agents, and Git hosting, but expresses concerns about UX (described as visual “horror”) and concrete privacy concerns when AI agents have access to sensitive conversations.
Resources #
Original Links #
- GitHub - block/buzz: A hive mind communication platform - Original link
Article reported and selected by the Human Technology eXcellence team processed through artificial intelligence (in this case with LLM HTX-EU-Claude-Haiku-4.5) on 2026-08-18 08:16 Original source: https://github.com/block/buzz
Related Articles #
- GitHub - different-ai/openwork: An open-source alternative to Claude Cowork, powered by OpenCode. - AI, Typescript, Open Source
- GitHub - rberg27/doom-coding: A guide on how to use your smartphone to code anywhere at any time. - Open Source
- GitHub - DGoettlich/history-llms: Information hub for our project training the largest possible historical language models. - AI, Go, Open Source