01What is Jack Dorsey's Buzz?
Jack Dorsey's Buzz is a new open-source project built on Nostra that works like your own version of Slack but for AI agents. It lets you chat with teammates and specialized agents in one shared place.
You can get access at buzz.xyz by downloading the app or opening the GitHub repo. The app is free to use and works with your existing Claude Code or Codex subscriptions, so you don't need extra API keys.
Buzz is designed to make it easy for your whole team to use AI agents without complicated setup. In this episode, Julian Goldie tests Buzz live, showing how it works step by step.
02Key Features of Jack Dorsey's Buzz
Buzz looks and feels a lot like Slack. You have channels for different topics, like design or engineering. You can tag agents to get things done, and each agent can have a specialty.
- Channels – Create public or private channels, and choose if they're ongoing or temporary.
- Agents – Use default agents (Honey, Bumble, Fizz) or create custom ones with specific purposes.
- Model flexibility – Switch between Claude Code and Codex (including GPT 5.6) as the underlying harness.
- Team access – Give your team access to AI agents easily, which is harder with other tools like Hermes or OpenClaw.
- Memory – Ask if a bug has been seen before; Buzz remembers.
- Branching rooms – Create branches for different conversations or tasks.
- Desktop and mobile apps – Available for desktop and coming for mobile.
You can also comment on shared files, like videos, and have a timer to see how long agents take to reply.
03How to Set Up Jack Dorsey's Buzz
Setting up Buzz is straightforward, especially if you use the app. Here’s the step-by-step process shown in the episode:
- Go to buzz.xyz and download the desktop app (or open the GitHub repo).
- Install the app on your computer.
- When you open it, choose to use Claude Code or Codex (you can install both).
- Sign in with your existing Claude or Codex account—no API key needed.
- Select your default harness (Julian chose Claude Code).
- Create a username and set up a community (or join an existing one).
- Create a channel (e.g., “content-ideas”) and choose its type and visibility.
- Invite team members and add agents to the channel.
Julian tried the manual GitHub setup with Docker and Hermit, but after 22 minutes it still wasn't done. He recommends using the app for a much faster setup.
04How to Use Buzz in Daily Work
Using Buzz is simple: you tag an agent in a channel and give it a task. In the test, Julian tagged Fizz to come up with content ideas about OpenClaw. Within seconds, Fizz replied with several ideas, including comparisons like OpenClaw versus Claude Code versus AutoGPT.
You can then ask the agent to create something. Julian asked Fizz to create a blog post, and it started writing a complete beginner’s guide to OpenClaw. It even used the custom workflows that Julian’s Claude setup already had for SEO blog posts.
- Tagging – You must tag the agent every time you want a reply. If you reply without tagging, it does nothing.
- Files and comments – You can comment on files like videos shared in Buzz.
- Agent customization – You can create agents from scratch, import a snapshot, choose the model (Codex, BuzzAgent, or Claude), and set environmental variables.
- Access control – Decide who on your team can talk to each agent.
Buzz also integrates with your local files and tools. In the test, it used computer users to open Firefox and check if the published link worked.
05Results and Real-World Test
In the episode, Julian tested Buzz thoroughly. The results were mostly positive, with a few hiccups.
- Agent response speed – Fizz replied to content requests quickly, and the timer showed response times.
- Blog post creation – Fizz created a full blog post and published it to WordPress. At first, the link seemed broken, but it turned out to be Julian’s cache. In an incognito tab, the post worked perfectly.
- Link hallucination – Initially, Fizz gave Julian a wrong link, but later it provided the correct one. Julian noted the mistake was his, not the agent’s.
- Cost – No extra API costs because Buzz uses your existing Claude Code or Codex subscriptions.
- Setup time – The app setup took a couple of minutes. Manual GitHub setup took over 22 minutes and wasn’t complete, so the app is the recommended route.
Julian concluded that Buzz is very good and definitely worth checking out, especially for teams that want to give members access to AI agents easily.
06Tips, Limits, and Mistakes to Avoid
Based on the episode, here are some practical tips and things to watch out for when using Buzz:
- Always tag agents – If you don't tag an agent, it won't respond. This is a key workflow requirement.
- Use the app for setup – The desktop app is much faster and easier than manual GitHub setup with Docker and Hermit.
- Check your cache – If a link seems broken, open it in an incognito tab before blaming the agent.
- Monitor token usage – If you run out of tokens on Claude, you can switch to Codex (GPT 5.6) without interruption.
- Be careful with links – Agents may hallucinate links; verify before relying on them.
One limitation is that Buzz is only useful if you use Claude or Codex, but for most users, that's a reasonable requirement. Also, the agent might need training for accuracy, as seen with the hallucinated link.
07Who Should Use Jack Dorsey's Buzz and Why
Buzz is for teams that want a simple, Slack-like way to collaborate with AI agents. If you're tired of configuring complex tools like Hermes or OpenClaw, Buzz offers a more intuitive experience.
It's especially useful if you already use Claude Code or Codex because you can leverage your existing subscriptions without extra costs. Julian points out that Slack got expensive for his team, and Buzz is free, making it a budget-friendly alternative.
Whether you need content ideas, blog posts, research, or coding help, Buzz lets you assign tasks to specialized agents and get results without deep technical know-how. It's a practical solution for teams looking to integrate AI into their daily workflow.
