'Buzz,' an Open-Source AI Collaboration Tool Co-Founded by Jack Dorsey, Emerges

Buzz is a platform specialized in collaboration between AI agents and humans, and it is being praised for being more user-friendly than other AI collaboration tools like Grokbot.

Buzz allows users to manage all subscribed AI models, such as Codex, Claude Code, and Grok, on a single screen. It can be easily linked using local subscription accounts without separate API integration, and users can collaborate with multiple people simultaneously, sharing work threads. This provides an efficient environment for team-based work utilizing various AI models.

Previously, when using multiple AI models simultaneously, it was difficult to remember which AI was doing what or maintain context. When a specific AI's usage ran out, a complex process was required to summarize the work as a prompt and hand it over to another AI. Buzz solves these difficulties in maintaining context and the complex prompt-sharing system during multi-AI tasks, enabling team-based AI work thread sharing.

Buzz leverages the nature of open-source, without a central operating company, by implementing a method of signing messages themselves. Each user and AI agent has a unique cryptographic key, and all messages, reactions, and commits are signed and recorded with that key. This ensures strong data integrity, making message forgery impossible even if the relay (server) is hacked.

Buzz's relay server only stores and transmits messages; it does not create them. Data enters sealed and leaves sealed, so the relay server cannot open or manipulate message content. For this, WSS (WebSocket Secure) connections are used to send and receive data, and a constantly running server environment, not a personal PC, is required for stable operation.

Buzz's relay server is written in Rust, and PostgreSQL is used for data storage. Real-time features are implemented through Redis, MinIO is utilized for file storage, and Git functionality is also built-in.

Buzz supports users in freely changing each AI backend model. You can switch between various models such as Codex, Claude Code, Grok, and LMS agents as needed. Context is maintained within a single thread, allowing for efficient model switching even in individual tasks.

If hosted on a local PC, real-time collaboration becomes impossible, so it is recommended to use an external server hosting service like Hostinger. The server should ideally have at least 2 cores and 8GB of RAM and can be used with a Code Factory discount coupon.

When operating a Buzz server, you must enter the Relay Owner Public Key (PubKey). This key is important information used for signing, and you can check and copy the private community key from the Buzz desktop app to use it. This key sets the server operator's permissions.

Therefore, you must prefix the URL with 'wss://' to successfully access the server and join the community.

Buzz offers an interface similar to Slack, visually displaying the collaboration process with AI agents. Users can give feedback on AI message confirmation using emoticons and check AI responses based on threads. Different backend models can be assigned to each AI agent to create an environment optimized for specific tasks.

For example, you can assign a specific purpose, such as 'YouTube Content Researcher,' and select and allocate an agent optimized for that role from various models like Claude Code, Codex, Cursor, and Grok Build.

In Buzz system settings, you can view the list of installed agents and expand new AI agent functionalities by adding runtimes. Additionally, you can activate 'Experimental features' to preview various experimental features, such as project functionality.

Buzz displays the real-time progress of AI agent tasks, similar to Slack's direct messages. You can check detailed information such as what research the AI is currently conducting and how long it has been working. This allows for efficient tracking of individual tasks in a mixed human-AI agent work environment within a channel.

This offers the advantage of continuously retaining previous context even when the agent model is changed. Therefore, it is highly useful for benchmarking tests or long-term project management based on the same workflow.

This allows users to manage work history in a UI similar to the familiar GitHub. Each commit history records which agent performed the work, signature information, and more, enabling transparent work management.

You can operate multiple community servers simultaneously through the Buzz desktop app. It's not limited to operating just one community; you can set up multiple servers as needed. Created communities can easily have other users join via an invite URL, providing a flexible collaboration environment.

Buzz is a Nostr-based service aiming to be a complete replacement for Slack. It specializes in record-keeping and LLM (Large Language Model) switching features, seeking to set a new standard for collaboration in the age of AI.