Jumpino
Updated
What changes when an agent joins the channel?
A separate chat window is easy to start and easy to lose. For a project, I want the brief, relevant files, decisions and verification to remain connected. A workspace can provide that shared context, but the team still needs to define what completion means.
The Buzz repository describes a self-hostable platform built around Nostr. Its architecture document places the relay at the centre of message storage and delivery. Clients and agents connect to that service. Signed events identify the author of an action; they do not prove that an AI answer is correct.
A small workflow worth trying
Start with one bounded task, such as reviewing the mobile contact page. Write down the expected result, the files the agent may change and how it should test the work. Give it access to the project it needs, then ask it to report the changed files and the evidence in the same conversation.
Use a review step before an action that affects other people. Reading a page, preparing a change and publishing it are different permissions. An agent should not infer permission to send a customer email just because it can read a customer record.
For a first trial, I would compare two complete runs of the same kind of task. Record how much clarification was needed, whether the result was correct and how long review took. Counting generated messages says little about useful work.
What should you check before adopting it?
| Question | What to establish |
|---|---|
| Who operates the relay? | Hosting, updates and responsibility for backups |
| What can each agent access? | Repository, channel and external tool permissions |
| Where does model inference run? | Provider settings and which data leaves the workspace |
| Who approves a release? | A named reviewer and a concrete verification step |
Self-hosting the workspace does not automatically make every connected model local. Read the deployment and provider settings separately. Keep secrets out of channel messages, and test removal of an agent's access as well as adding it.
Is Buzz a replacement for project management?
It can bring conversations and agent work closer together. It cannot choose priorities or judge business trade-offs for the team. The agent vision is useful context, but distinguish planned behaviour from features present in your installed version.
If your real need is an assistant inside a business workflow, begin with the task and its review rules. See how I approach custom AI integrations.