tui: handle multi part responses

pkg/tui/tui.go

@@ -33,7 +33,8 @@ type model struct {
 	conversation    *models.Conversation
 	messages        []models.Message
 	waitingForReply bool
-	replyChan       chan string
+	replyChan       chan models.Message
+	replyChunkChan  chan string
 	replyCancelFunc context.CancelFunc
 	err             error
 
@@ -57,6 +58,8 @@ type (
 	msgResponseChunk string
 	// sent when response is finished being received
 	msgResponseEnd string
+	// sent on each completed reply
+	msgReply models.Message
 	// sent when a conversation is (re)loaded
 	msgConversationLoaded *models.Conversation
 	// send when a conversation's messages are laoded
@@ -80,7 +83,8 @@ func (m model) Init() tea.Cmd {
 	return tea.Batch(
 		textarea.Blink,
 		m.loadConversation(m.convShortname),
-		waitForChunk(m.replyChan),
+		m.waitForChunk(),
+		m.waitForReply(),
 	)
 }
 
@@ -135,7 +139,21 @@ func (m model) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
 			})
 		}
 		m.updateContent()
-		cmd = waitForChunk(m.replyChan) // wait for the next chunk
+		cmd = m.waitForChunk() // wait for the next chunk
+	case msgReply:
+		// the last reply that was being worked on is finished
+		reply := models.Message(msg)
+		last := len(m.messages) - 1
+		if last < 0 {
+			panic("Unexpected messages length handling msgReply")
+		}
+		if reply.Role == models.MessageRoleToolCall && m.messages[last].Role == models.MessageRoleAssistant {
+			m.messages[last] = reply
+		} else if reply.Role != models.MessageRoleAssistant {
+			m.messages = append(m.messages, reply)
+		}
+		m.updateContent()
+		cmd = m.waitForReply()
 	case msgResponseEnd:
 		m.replyCancelFunc = nil
 		m.waitingForReply = false
@@ -173,7 +191,8 @@ func initialModel(ctx *lmcli.Context, convShortname string) model {
 		ctx:           ctx,
 		convShortname: convShortname,
 
-		replyChan: make(chan string),
+		replyChan:      make(chan models.Message),
+		replyChunkChan: make(chan string),
 	}
 
 	m.content = viewport.New(0, 0)
@@ -253,9 +272,15 @@ func (m *model) loadMessages(c *models.Conversation) tea.Cmd {
 	}
 }
 
-func waitForChunk(ch chan string) tea.Cmd {
+func (m *model) waitForReply() tea.Cmd {
 	return func() tea.Msg {
-		return msgResponseChunk(<-ch)
+		return msgReply(<-m.replyChan)
+	}
+}
+
+func (m *model) waitForChunk() tea.Cmd {
+	return func() tea.Msg {
+		return msgResponseChunk(<-m.replyChunkChan)
 	}
 }
 
@@ -281,12 +306,16 @@ func (m *model) promptLLM() tea.Cmd {
 			ToolBag:     toolBag,
 		}
 
+		replyHandler := func(msg models.Message) {
+			m.replyChan <- msg
+		}
+
 		ctx, replyCancelFunc := context.WithCancel(context.Background())
 		m.replyCancelFunc = replyCancelFunc
 
-		// TODO: supply a reply callback and handle error
+		// TODO: handle error
 		resp, _ := completionProvider.CreateChatCompletionStream(
-			ctx, requestParams, m.messages, nil, m.replyChan,
+			ctx, requestParams, m.messages, replyHandler, m.replyChunkChan,
 		)
 
 		return msgResponseEnd(resp)