diff --git a/crates/goose-server/src/routes/reply.rs b/crates/goose-server/src/routes/reply.rs
index 5b98f9c701..eced7bc7bb 100644
--- a/crates/goose-server/src/routes/reply.rs
+++ b/crates/goose-server/src/routes/reply.rs
@@ -75,7 +75,9 @@ fn track_tool_telemetry(content: &MessageContent, all_messages: &[Message]) {
 
 #[derive(Debug, Deserialize, Serialize, utoipa::ToSchema)]
 pub struct ChatRequest {
-    messages: Vec<Message>,
+    user_message: Message,
+    #[serde(default)]
+    conversation_so_far: Option<Vec<Message>>,
     session_id: String,
     recipe_name: Option<String>,
     recipe_version: Option<String>,
@@ -233,7 +235,8 @@ pub async fn reply(
     let stream = ReceiverStream::new(rx);
     let cancel_token = CancellationToken::new();
 
-    let messages = Conversation::new_unvalidated(request.messages);
+    let user_message = request.user_message;
+    let conversation_so_far = request.conversation_so_far;
 
     let task_cancel = cancel_token.clone();
     let task_tx = tx.clone();
@@ -255,7 +258,7 @@ pub async fn reply(
             }
         };
 
-        let session = match SessionManager::get_session(&session_id, false).await {
+        let session = match SessionManager::get_session(&session_id, true).await {
             Ok(metadata) => metadata,
             Err(e) => {
                 tracing::error!("Failed to read session for {}: {}", session_id, e);
@@ -278,20 +281,21 @@ pub async fn reply(
             retry_config: None,
         };
 
-        let user_message = match messages.last() {
-            Some(msg) => msg,
-            _ => {
-                let _ = stream_event(
-                    MessageEvent::Error {
-                        error: "Reply started with empty messages".to_string(),
-                    },
-                    &task_tx,
-                    &task_cancel,
-                )
-                .await;
-                return;
+        let mut all_messages = match conversation_so_far {
+            Some(history) => {
+                let conv = Conversation::new_unvalidated(history);
+                if let Err(e) = SessionManager::replace_conversation(&session_id, &conv).await {
+                    tracing::warn!(
+                        "Failed to replace session conversation for {}: {}",
+                        session_id,
+                        e
+                    );
+                }
+                conv
             }
+            None => session.conversation.unwrap_or_default(),
         };
+        all_messages.push(user_message.clone());
 
         let mut stream = match agent
             .reply(
@@ -316,8 +320,6 @@ pub async fn reply(
             }
         };
 
-        let mut all_messages = messages.clone();
-
         let mut heartbeat_interval = tokio::time::interval(Duration::from_millis(500));
         loop {
             tokio::select! {
@@ -478,7 +480,8 @@ mod tests {
                 .header("x-secret-key", "test-secret")
                 .body(Body::from(
                     serde_json::to_string(&ChatRequest {
-                        messages: vec![Message::user().with_text("test message")],
+                        user_message: Message::user().with_text("test message"),
+                        conversation_so_far: None,
                         session_id: "test-session".to_string(),
                         recipe_name: None,
                         recipe_version: None,
diff --git a/ui/desktop/openapi.json b/ui/desktop/openapi.json
index 4d0e0a6342..ca63e46a53 100644
--- a/ui/desktop/openapi.json
+++ b/ui/desktop/openapi.json
@@ -2715,15 +2715,16 @@
       "ChatRequest": {
         "type": "object",
         "required": [
-          "messages",
+          "user_message",
           "session_id"
         ],
         "properties": {
-          "messages": {
+          "conversation_so_far": {
             "type": "array",
             "items": {
               "$ref": "#/components/schemas/Message"
-            }
+            },
+            "nullable": true
           },
           "recipe_name": {
             "type": "string",
@@ -2735,6 +2736,9 @@
           },
           "session_id": {
             "type": "string"
+          },
+          "user_message": {
+            "$ref": "#/components/schemas/Message"
           }
         }
       },
diff --git a/ui/desktop/src/api/types.gen.ts b/ui/desktop/src/api/types.gen.ts
index 0166872c13..ea60af4a5a 100644
--- a/ui/desktop/src/api/types.gen.ts
+++ b/ui/desktop/src/api/types.gen.ts
@@ -59,10 +59,11 @@ export type CallToolResponse = {
 };
 
 export type ChatRequest = {
-    messages: Array<Message>;
+    conversation_so_far?: Array<Message> | null;
     recipe_name?: string | null;
     recipe_version?: string | null;
     session_id: string;
+    user_message: Message;
 };
 
 export type CheckProviderRequest = {
diff --git a/ui/desktop/src/hooks/useChatStream.ts b/ui/desktop/src/hooks/useChatStream.ts
index 01555361b5..41580f7609 100644
--- a/ui/desktop/src/hooks/useChatStream.ts
+++ b/ui/desktop/src/hooks/useChatStream.ts
@@ -292,12 +292,13 @@ export function useChatStream({
         window.dispatchEvent(new CustomEvent('session-created'));
       }
 
-      // Build message list: add new message if provided, otherwise continue with existing
+      const newMessage = hasNewMessage
+        ? createUserMessage(userMessage)
+        : messagesRef.current[messagesRef.current.length - 1];
       const currentMessages = hasNewMessage
-        ? [...messagesRef.current, createUserMessage(userMessage)]
+        ? [...messagesRef.current, newMessage]
         : [...messagesRef.current];
 
-      // Update UI with new message before streaming
       if (hasNewMessage) {
         updateMessages(currentMessages);
       }
@@ -310,7 +311,7 @@ export function useChatStream({
         const { stream } = await reply({
           body: {
             session_id: sessionId,
-            messages: currentMessages,
+            user_message: newMessage,
           },
           throwOnError: true,
           signal: abortControllerRef.current.signal,
@@ -356,7 +357,7 @@ export function useChatStream({
         const { stream } = await reply({
           body: {
             session_id: sessionId,
-            messages: currentMessages,
+            user_message: responseMessage,
           },
           throwOnError: true,
           signal: abortControllerRef.current.signal,