Guide

Multiplayer chat

A Space's chat is a single thread that everyone — humans and AI — writes into. The two rules that make it not chaos: AI is silent until @mentioned, and presence is opt-in not push.

The two rules

Silence by default. No AI replies unless someone types @claude, @gpt-5, or another agent name. Plain messages stay plain. This is the A6 invariant — see the architecture docs if you want the formal version.

No push notifications. Teammates see live updates only when the Space is open in a browser tab. If you need someone's attention right now, use email or your existing tools.

Typical flow

Open a Space. The thread is the room.
Type a message — to a teammate, no @mention. They see it live (typing indicator + new bubble).
Mention an agent when you want the AI to chime in. @claude what should we test before launch?
Mention multiple agents to chain. @github @claude what issues are blocking? calls GitHub first, then has Claude summarise the result. Cap of 3 MCPs per message.
Save what you decide. Tell the agent "remember for the team that …" and it calls add_memory. Future agents in any Space will see it.

Live indicators

Typing indicators show below the composer when teammates are mid-message.
Online presence appears in the Participants rail. We don't broadcast away/idle states; if a tab is open, you're considered online.
Streaming AI renders incrementally so you can interrupt or react before the agent finishes.

Threads + reactions

Hover any message to see Reply in thread + reaction picker. Threads keep tangents out of the main flow. Reactions render inline with counts; click to toggle yours.

What this isn't

Switchy chat is not a Slack replacement. It optimises for shared context with AI in the loop, not 1:1 messaging or org-wide announcements. If your team needs DMs and channel pings, keep using Slack and let Switchy handle the AI-augmented work.