Apple/Web Streaming Parity

Status: Linux static audit
Created: 2026-06-01
Scope: AI task events, typing state, chunks, thinking, post-processing, queued messages

Goal

Apple streaming must match web-visible AI response behavior: task initiation, typing indicator, incremental content chunks, thinking sections, final message readiness, post-processing metadata, suggestions, and queued-message handling.

Web Source

• frontend/packages/ui/src/services/websocketService.ts
• frontend/packages/ui/src/services/chatSyncServiceHandlersAI.ts
• frontend/packages/ui/src/services/chatSyncService.ts
• frontend/packages/ui/src/components/ChatMessage.svelte
• frontend/packages/ui/src/components/ThinkingSection.svelte
• Web specs that assert message-assistant, typing-indicator, and streaming completion across skill flows

Apple Source

• apple/OpenMates/Sources/Core/Networking/WebSocketManager.swift
• apple/OpenMates/Sources/Core/Networking/StreamingClient.swift
• apple/OpenMates/Sources/Features/Chat/ViewModels/ChatViewModel.swift
• apple/OpenMates/Sources/Features/Chat/Views/ChatView.swift
• apple/OpenMates/Sources/Features/Chat/Views/ThinkingSectionView.swift
• apple/OpenMates/Sources/Features/Chat/Views/ProcessingDetailsView.swift

Confirmed Apple Behavior From Source

• WebSocketManager routes ai_task_initiated to StreamingClient.StreamEvent.taskInitiated.
• ai_typing_started is routed with chat metadata, including title, icons, category, model/provider/region, user message ID, and encrypted chat key.
• ai_message_update is routed as chunk events with sequence, full content so far, final flag, user message ID, category, model name, and rejection reason.
• thinking_chunk and thinking_complete have dedicated stream events.
• ai_message_ready has a dedicated stream event.
• preprocessing_step and post_processing_completed route to stream events.
• StreamingClient buffers up to 80 events per chat when no active stream exists, which is useful for race conditions where events arrive before a view subscribes.
• WebSocket routes message_queued through .wsMessageReceived but not into StreamingClient directly.

Web Parity Risks

• Web has explicit queued ai_typing_started handling for a missing chat shell (OPE-360). Apple buffering is per chat ID and likely helps, but parity depends on ChatViewModel subscribing at the correct time and creating the shell from new_chat_message or metadata.
• Web handles background chat completion notifications and unread increments in AI handlers. Apple streaming source alone does not prove equivalent background notification behavior.
• Web ai_message_update comments show extensive debug handling around streaming. Apple route is simpler; verify final chunk ordering, duplicate chunks, and missed final events.
• message_queued should surface in the composer/input area on web, not as a generic notification. Apple needs equivalent UI behavior if supported.
• post_processing_completed carries follow-up suggestions, new chat suggestions, summary, tags, and updated title. Apple must confirm these are applied to UI and persisted.

Testability IDs Needed

• typing-indicator
• message-assistant
• thinking-section
• processing-details
• follow-up-suggestions
• new-chat-suggestions
• message-queued-banner or native equivalent

Mac Verification Checklist

• Send a prompt and verify typing indicator appears before first chunk.
• Verify chunks update the same assistant bubble rather than duplicating messages.
• Verify thinking chunks appear and collapse/complete like web.
• Verify final response persists and no streaming indicator remains stuck.
• Trigger post-processing and verify follow-up suggestions, summary/title updates, and new-chat suggestions.
• Trigger an active-task queued message and verify it appears in the composer/input surface, not as an unrelated notification.
• Start streaming on web and open Apple mid-stream to verify buffered event handling.

First Implementation Tasks

• Audit ChatViewModel event handling against every StreamingClient.StreamEvent case.
• Add stable identifiers for typing, thinking, processing, and suggestions UI.
• Add a streaming race test plan for chat shell creation before/after ai_typing_started.