Fix history loading race condition in ThreadManager

assistant/src/__tests__/__snapshots__/ipc-snapshot.test.ts.snap

@@ -529,6 +529,7 @@ exports[`IPC message snapshots ServerMessage types history_response serializes t
       "timestamp": 1700000001,
     },
   ],
+  "sessionId": "sess-history-001",
   "type": "history_response",
 }
 `;

assistant/src/__tests__/ipc-snapshot.test.ts

@@ -336,6 +336,7 @@ const serverMessages: Record<ServerMessageType, ServerMessage> = {
   },
   history_response: {
     type: 'history_response',
+    sessionId: 'sess-history-001',
     messages: [
       { role: 'user', text: 'Hello', timestamp: 1700000000 },
       { role: 'assistant', text: 'Hi there!', timestamp: 1700000001 },

assistant/src/daemon/handlers.ts

@@ -673,7 +673,7 @@ function handleHistoryRequest(
     timestamp: m.timestamp,
     ...(m.toolCalls.length > 0 ? { toolCalls: m.toolCalls } : {}),
   }));
-  ctx.send(socket, { type: 'history_response', messages: historyMessages });
+  ctx.send(socket, { type: 'history_response', sessionId: msg.sessionId, messages: historyMessages });
 }
 
 export function mergeToolResults(messages: ParsedHistoryMessage[]): ParsedHistoryMessage[] {

assistant/src/daemon/ipc-protocol.ts

@@ -533,6 +533,7 @@ export interface HistoryResponseToolCall {
 
 export interface HistoryResponse {
   type: 'history_response';
+  sessionId: string;
   messages: Array<{
     role: string;
     text: string;

clients/macos/vellum-assistant/Features/Chat/ChatViewModel.swift

@@ -968,7 +968,15 @@ final class ChatViewModel: ObservableObject {
     }
 
     /// Populate messages from history data returned by the daemon.
+    /// Only replaces messages if the user hasn't sent any new messages yet,
+    /// preventing a late history_response from overwriting live conversation.
     func populateFromHistory(_ historyMessages: [HistoryResponseMessage.HistoryMessageItem]) {
+        let hasUserSentMessages = messages.contains { $0.role == .user }
+        if hasUserSentMessages {
+            isHistoryLoaded = true
+            return
+        }
+
         var chatMessages: [ChatMessage] = []
         for item in historyMessages {
             let role: ChatRole = item.role == "assistant" ? .assistant : .user

clients/macos/vellum-assistant/Features/MainWindow/ThreadManager.swift

@@ -19,8 +19,9 @@ final class ThreadManager: ObservableObject {
     private var viewModelCancellable: AnyCancellable?
     private var connectionCancellable: AnyCancellable?
 
-    /// Tracks which thread is currently awaiting a history_response from the daemon.
-    private var pendingHistoryThreadId: UUID?
+    /// Maps session IDs to thread IDs for in-flight history_request messages,
+    /// so rapid tab switches don't cause history from one thread to land in another.
+    private var pendingHistoryBySessionId: [String: UUID] = [:]
 
     /// Called when an inline confirmation response should dismiss the floating panel.
     var confirmationDismissHandler: ((String) -> Void)?
@@ -175,23 +176,22 @@ final class ThreadManager: ObservableObject {
     private func loadHistoryForActiveThreadIfNeeded() {
         guard let activeThreadId else { return }
         guard let thread = threads.first(where: { $0.id == activeThreadId }) else { return }
-        guard thread.sessionId != nil else { return }
+        guard let sessionId = thread.sessionId else { return }
         guard let viewModel = chatViewModels[activeThreadId] else { return }
         guard !viewModel.isHistoryLoaded else { return }
 
-        pendingHistoryThreadId = activeThreadId
+        pendingHistoryBySessionId[sessionId] = activeThreadId
 
         do {
-            try daemonClient.sendHistoryRequest(sessionId: thread.sessionId!)
+            try daemonClient.sendHistoryRequest(sessionId: sessionId)
         } catch {
             log.error("Failed to send history_request: \(error.localizedDescription)")
-            pendingHistoryThreadId = nil
+            pendingHistoryBySessionId.removeValue(forKey: sessionId)
         }
     }
 
     private func handleHistoryResponse(_ response: HistoryResponseMessage) {
-        guard let threadId = pendingHistoryThreadId else { return }
-        pendingHistoryThreadId = nil
+        guard let threadId = pendingHistoryBySessionId.removeValue(forKey: response.sessionId) else { return }
 
         guard let viewModel = chatViewModels[threadId] else { return }
         viewModel.populateFromHistory(response.messages)

clients/macos/vellum-assistant/IPC/IPCMessages.swift

@@ -691,6 +691,7 @@ struct SessionListResponseMessage: Decodable, Sendable {
 /// Response containing message history for a session.
 /// Wire type: `"history_response"`
 struct HistoryResponseMessage: Decodable, Sendable {
+    let sessionId: String
     struct HistoryToolCallItem: Decodable, Sendable {
         let name: String
         let input: [String: AnyCodable]