diff --git a/pkg/tui/tui.go b/pkg/tui/tui.go
index de27cbe..c9df9c5 100644
--- a/pkg/tui/tui.go
+++ b/pkg/tui/tui.go
@@ -39,6 +39,7 @@ type model struct {
 	replyChunkChan  chan string
 	replyCancelFunc context.CancelFunc
 	err             error
+	persistence     bool // whether we will save new messages in the conversation
 
 	// ui state
 	focus          focusState
@@ -81,7 +82,6 @@ var (
 	contentStyle = lipgloss.NewStyle().
 			Padding(1)
 	footerStyle = lipgloss.NewStyle().
-			Faint(true).
 			BorderTop(true).
 			BorderStyle(lipgloss.NormalBorder())
 )
@@ -95,8 +95,14 @@ func (m model) Init() tea.Cmd {
 	)
 }
 
+func wrapError(err error) tea.Cmd {
+	return func() tea.Msg {
+		return msgError(err)
+	}
+}
+
 func (m model) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
-	var cmd tea.Cmd
+	var cmds []tea.Cmd
 
 	switch msg := msg.(type) {
 	case tea.KeyMsg:
@@ -107,6 +113,8 @@ func (m model) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
 			} else {
 				return m, tea.Quit
 			}
+		case "ctrl+p":
+			m.persistence = !m.persistence
 		case "q":
 			if m.focus != focusInput {
 				return m, tea.Quit
@@ -130,7 +138,7 @@ func (m model) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
 		m.updateContent()
 	case msgConversationLoaded:
 		m.conversation = (*models.Conversation)(msg)
-		cmd = m.loadMessages(m.conversation)
+		cmds = append(cmds, m.loadMessages(m.conversation))
 	case msgMessagesLoaded:
 		m.setMessages(msg)
 		m.updateContent()
@@ -146,7 +154,7 @@ func (m model) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
 			})
 		}
 		m.updateContent()
-		cmd = m.waitForChunk() // wait for the next chunk
+		cmds = append(cmds, m.waitForChunk()) // wait for the next chunk
 	case msgReply:
 		// the last reply that was being worked on is finished
 		reply := models.Message(msg)
@@ -159,18 +167,24 @@ func (m model) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
 		} else if reply.Role != models.MessageRoleAssistant {
 			m.addMessage(reply)
 		}
+
+		if m.persistence && m.conversation != nil && m.conversation.ID > 0 {
+			cmds = append(cmds, m.persistRecentMessages())
+		}
+
 		m.updateContent()
-		cmd = m.waitForReply()
+		cmds = append(cmds, m.waitForReply())
 	case msgResponseEnd:
 		m.replyCancelFunc = nil
 		m.waitingForReply = false
 		m.status = "Press ctrl+s to send"
 	}
 
-	if cmd != nil {
-		return m, cmd
+	if len(cmds) > 0 {
+		return m, tea.Batch(cmds...)
 	}
 
+	var cmd tea.Cmd
 	m.input, cmd = m.input.Update(msg)
 	if cmd != nil {
 		return m, cmd
@@ -225,10 +239,18 @@ func (m *model) inputView() string {
 }
 
 func (m *model) footerView() string {
-	segmentStyle := lipgloss.NewStyle().PaddingLeft(1).PaddingRight(1)
+	segmentStyle := lipgloss.NewStyle().PaddingLeft(1).PaddingRight(1).Faint(true)
 	segmentSeparator := "|"
 
+	saving := ""
+	if m.persistence {
+		saving = segmentStyle.Copy().Bold(true).Foreground(lipgloss.Color("2")).Render("✅💾")
+	} else {
+		saving = segmentStyle.Copy().Bold(true).Foreground(lipgloss.Color("1")).Render("❌💾")
+	}
+
 	leftSegments := []string{
+		saving,
 		segmentStyle.Render(m.status),
 	}
 	rightSegments := []string{
@@ -258,6 +280,7 @@ func initialModel(ctx *lmcli.Context, convShortname string) model {
 	m := model{
 		ctx:           ctx,
 		convShortname: convShortname,
+		persistence:    true,
 
 		replyChan:      make(chan models.Message),
 		replyChunkChan: make(chan string),
@@ -298,11 +321,29 @@ func (m *model) handleInputKey(msg tea.KeyMsg) tea.Cmd {
 		if strings.TrimSpace(userInput) == "" {
 			return nil
 		}
-		m.input.SetValue("")
-		m.addMessage(models.Message{
+
+		reply := models.Message{
 			Role:    models.MessageRoleUser,
 			Content: userInput,
-		})
+		}
+
+		if m.persistence && m.conversation != nil && m.conversation.ID > 0 {
+			// ensure all messages up to the one we're about to add are
+			// persistent
+			cmd := m.persistRecentMessages()
+			if cmd != nil {
+				return cmd
+			}
+			// persist our new message, returning with any possible errors
+			savedReply, err := m.ctx.Store.AddReply(m.conversation, reply)
+			if err != nil {
+				return wrapError(err)
+			}
+			reply = *savedReply
+		}
+
+		m.input.SetValue("")
+		m.addMessage(reply)
 
 		m.updateContent()
 		m.content.GotoBottom()
@@ -382,6 +423,20 @@ func (m *model) promptLLM() tea.Cmd {
 	}
 }
 
+func (m *model) persistRecentMessages() tea.Cmd {
+	for i, msg := range m.messages {
+		if msg.ID > 0 {
+			continue
+		}
+		newMessage, err := m.ctx.Store.AddReply(m.conversation, msg)
+		if err != nil {
+			return wrapError(err)
+		}
+		m.setMessage(i, *newMessage)
+	}
+	return nil
+}
+
 func (m *model) setMessages(messages []models.Message) {
 	m.messages = messages
 	m.highlightCache = make([]string, len(messages))
@@ -436,11 +491,16 @@ func (m *model) updateContent() {
 			icon = "🔧"
 		}
 
+		var saved string
+		if message.ID == 0 {
+			saved = lipgloss.NewStyle().Faint(true).Render(" (not saved)")
+		}
+
 		// write message heading with space for content
-		header := fmt.Sprintf("%s\n\n", style.Render(icon+friendly))
+		header := fmt.Sprintf("%s\n\n", style.Render(icon+friendly)+saved)
 		sb.WriteString(header)
 
-		// TODO: render something for tool calls/results?
+		// TODO: special rendering for tool calls/results?
 
 		// write message contents
 		var highlighted string