Chat Events (contract)

Your backend can emit the defaults below or advertise/rename them via mapping or discovery. The TypeScript payload types live in your project as ChatEvents.

Server → Client (default names)

• chat:message → { chatId, messageId, userId, text, createdAt }
• ai:processing → { chatId, status: "queued"|"working"|"retrying", etaMs?, reason?, requestId? }
• ai:token → { chatId, token, index, done?: false, requestId? }
• ai:message → { chatId, messageId, role:"assistant", text, createdAt, usage?, options?, requestId? }
• ai:error → { chatId, code?, message, details?, requestId? }
• presence:update → { chatId, onlineUserIds: Array<string|number> }

Client → Server (default names)

• user:message → { chatId, messageId, userId, text, parts?, requestId? }
• user:typingStart / user:typingStop → { chatId, userId }
• ai:abort → { chatId, reason? }
• chat:read → { chatId, userId, messageIds, readAt }

requestId is critical for correlating a user turn to the token/final events.

Remapping: If your server uses e.g. llm:final for ai:message, set eventNames: { AI_MESSAGE: "llm:final" } or provide an eventResolver.

Updates (additive)

• chatId is optional when your backend doesn’t use rooms.
• Any meta provided in AIChatSocketOptions.meta is automatically merged into client→server emits (user:message, user:typing*, ai:abort, chat:read).