Buzz, first attempt

I’m currently working on setting up Buzz to have one unified workspace for my agents and to have them collaborate across hosts. It’s WIP live at https://buzz.harmont.xyz*

*in browser link will only show you repos, but you can also use the address to connect to the relay using Buzz Desktop, phone-app or CLI.

Intended architecture

I initially envisioned something like this:

Host 1 (laptop):

• agent1: running an independant agent-process with buzz-acp using opencode as harness (starting when laptop is awake)
• desktop-app started on demand logged in as me, Filip (human user)

Host 2 (PC):

• agent2: running an independant agent-process with buzz-acp using opencode as harness (starting when PC is on)
• desktop-app (starting when PC is on) with…
  • Filip (the same human user)
  • Fizz, Honey and Pollen (Buzz default agents)
  • Agent3: “Hermato”, using buzz built-in ACP connection to Hermes.

Phone-app logged in as Filip (the same human user)

Current state and blockers

Default agents and shared human identities

The desktop-app auto-creates default agents, but it creates separate agent-identities even if the desktop-app is running with the same user ID. Meaning that you’ll have multiple instances of the default-agents.

Projects

Projects are currently experimental so these things are to expect. But the issue-tracking is VERY limited in desktop UI and also for agents to change state.

Limited power of multi-host set-up

Buzz has built-in for you to share your hardware to run local models on to the network. That’s good. But for sharing agents the current view is that a human has one computer, because even though you can log in from multiple desktops you can’t control the agent processes on a remote machine.

To be continued…