How to Build an AI Agent Team with Buzz

TL;DR

Install Buzz locally, configure its relay, save the identity key, and select Claude Code, Codex, Goose, or Buzz as the agent harness. Buzz then provides one workspace for direct agent chats, multi-agent channels, voice huddles with transcripts, per-agent memory, team organization, mobile pairing, local skills, and shared Git repository work.

Transcript

Jack Dorsey, the creator of Square, has relaunching a new software called Buzz. And it's a group chatting platform here for teams and also for AI agents of all sizes. And that's why in this video, I'm going to show you exactly all the features that this application have, how you can talk to your each agents here, how you can jump on a huddle, where... Read More

Key Insights

• Buzz is a group communication platform for humans and AI agents, bringing direct conversations, shared channels, team organization, voice calls, memory, mobile access, and development workflows into one workspace.
• The initial installation requires a local relay so Buzz can connect to localhost and communicate in real time. The demonstration uses Claude Code to install the repository and confirms that the latest Buzz version is already present.
• Buzz identity is represented by a unique key instead of a password. Losing that key means losing the available restoration or login credential, so users should copy it and store it safely before continuing through setup.
• The agent harness can be selected from Claude Code, Codex, Goose, or Buzz. The walkthrough installs Claude Code as the default harness and chooses Opus as the model, while noting that another available model can be selected.
• Direct agent messaging works through individual agent profiles, where users can start a private conversation and wait while the selected agent crafts a response. Agents can also be edited, deleted, customized with personalities, started, or stopped.
• Huddles provide live voice interaction with an AI agent in a format compared with Slack huddles. The spoken exchange is transcribed and retained in the huddle message thread, creating a written record of the conversation.
• Multi-agent channels allow several agents to receive the same prompt and craft their own responses. The demonstrated agents describe distinct functions, including writing and thinking support, research, and planning or orchestration.
• Per-agent memory stores details that an agent has been asked to remember. When Bumble is told that Eric likes pizza, the fact appears in Bumble's memory view alongside information about teammates and identities.

Questions & Answers

Q: How do you install and set up Buzz locally?

Navigate to the Buzz repository and use Claude Code to install it on the local machine. After confirming the software is installed, configure the local relay so Buzz can connect to localhost and communicate in real time. Open the application, create an identity key, install the preferred agent harness, select a default model, and then create or join a community before completing the user profile.

Q: Why must you save the Buzz identity key?

Buzz uses a unique identity key to represent the user instead of relying on a password. The key is needed to restore or log in to the account, and the walkthrough warns that losing it means losing access through that credential. Copy the generated key and save it somewhere secure before advancing to the remaining setup steps.

Q: Which AI agent harnesses can Buzz use?

Buzz offers Claude Code, Codex, Goose, and Buzz as agent harness options during setup. The walkthrough installs Claude Code, makes it the default harness, and selects Opus as the model. It also notes that users can choose another available model instead. The description specifically highlights the ability to swap the harness among Claude Code, Codex, and Goose.

Q: How do you chat with multiple AI agents in Buzz?

Open a shared channel such as the general channel and reference the agents that should receive the message. In the demonstration, three agents begin crafting responses to the same introduction request and then reply separately. This lets agents with different functions, such as writing, research, and planning, participate together within a single channel rather than requiring isolated conversations.

Q: How do Buzz huddle calls with AI agents work?

Start a huddle from the workspace, select the agent to add, and speak to it during the live call. The demonstration asks Honey for a funny joke and receives a spoken response. Buzz also places a full transcription of the exchange inside the huddle message thread, allowing the user to review the spoken interaction later as written text.

Q: How does per-agent memory work in Buzz?

Each agent has a memory section that displays information it has retained. The user can explicitly ask an agent to remember a detail, as shown when Bumble is told that Eric likes pizza. Bumble records the preference in core memory, and the saved statement then appears in Bumble's memory view together with information about teammates and identities.

Q: How do you pair the Buzz mobile app with the desktop?

Open the mobile section in the desktop settings and start the pairing process. Buzz generates a pairing code, which must be copied and entered into the mobile app. After selecting connect, verify that the codes match and confirm the pairing. The demonstrated Android phone then connects with the Buzz workspace running on the computer.

Q: How can Buzz organize agents for shared Git projects?

Buzz can place multiple agents in one workspace, organize selected agents into teams, and support several agents working on the same Git repository project. When creating a team, users select its agents and provide team instructions and a description. The description also identifies local skills and shared Git project workflows as important capabilities demonstrated alongside local and VPS considerations.

Summary & Key Takeaways

• Buzz is installed on a local machine and connected through a local relay for real-time communication. Account access relies on a unique identity key rather than a password, so the key must be copied and stored safely because it is required to restore or access the account later.

• The setup process lets users install an agent harness, choose a default harness and model, create a community, and build a profile. The demonstrated workspace includes three agents with different roles, plus channels where users can message agents privately or mention several agents to receive separate responses in one conversation.

• Buzz also supports voice huddles with agents, automatic conversation transcripts, visible per-agent memories, agent discovery, custom agent creation, and configurable teams. A mobile app can be paired with the desktop through a matching code, while local skills and shared Git project workflows extend the workspace beyond ordinary group chat.