Skip to content

04-chat-architecture.md

Latest commit

 

History

History
32 lines (24 loc) · 1.61 KB

04-chat-architecture.md

File metadata and controls

32 lines (24 loc) · 1.61 KB
Raw

Chat Architecture

Runtime topology

AppModel.conversationRuntimes
  ConversationKey A → PiConversationModel A → PiRPCClient A → pi process A
  ConversationKey B → PiConversationModel B → PiRPCClient B → pi process B
  • AppModel.swift lazily creates one ConversationRuntime per selected or pending chat.
  • activeConversationModel is only the center view's pointer to the selected runtime; it is not the sole live conversation.
  • Combine subscriptions mirror each model's running/loading/error state into conversationRuntimeStates for sidebar indicators.
  • Keyed callbacks persist session path, title summary, and transcript to the owning chat even after the user navigates away.
  • Switching chats does not stop the old runtime; archive/promotion cleanup and app termination do.

One conversation model

PiConversationModel.swift owns:

  • Draft and attachment state.
  • Cached/live transcript items.
  • Session readiness, loading, failure, and running state.
  • Current model and thinking level.
  • Pending initial prompt and current Pi session path.
  • One PiRPCClient, scoped to a working directory and planning mode.

Project chat vs Quick Chat

  • Project chat working directory is the project folder and tools are enabled.
  • Quick Chat working directory is the user's home folder, Pi launches with --no-tools, and prompts receive an explicit planning-only preamble.
  • Promotion creates a new project chat/runtime; it does not retarget the Quick Chat's process.