feat(tui): show router inference metrics beside router lines

apps/tui-go/internal/app/demo.go

@@ -37,9 +37,9 @@ func PreviewFrame(kind string, w, h int) string {
 		m.sessionEntered = true
 		m.routerConnected = true
 		m.routerMessages["04a546aa"] = []RouterEntry{
-			{"user", "write a healthcheck script for the api"},
-			{"router", "I'll create a script that pings /health and checks the status code, then writes the results to a timestamped file."},
-			{"tool", "--- a/healthcheck.sh\n+++ b/healthcheck.sh\n@@ -0,0 +1,4 @@\n+#!/usr/bin/env bash\n+curl -sf http://localhost:8080/health\n+echo \"ok $(date)\"\n"},
+			{Role: "user", Content: "write a healthcheck script for the api"},
+			{Role: "router", Content: "I'll create a script that pings /health and checks the status code, then writes the results to a timestamped file."},
+			{Role: "tool", Content: "--- a/healthcheck.sh\n+++ b/healthcheck.sh\n@@ -0,0 +1,4 @@\n+#!/usr/bin/env bash\n+curl -sf http://localhost:8080/health\n+echo \"ok $(date)\"\n"},
 		}
 		if s := m.session("04a546aa"); s != nil {
 			s.CurrentStage = "write_script"
@@ -127,7 +127,7 @@ func PreviewFrame(kind string, w, h int) string {
 		m.selectedID = "04a546aa"
 		m.sessionEntered = true
 		m.routerMessages["04a546aa"] = []RouterEntry{
-			{"tool", "--- a/healthcheck.sh\n+++ b/healthcheck.sh\n@@ -1,2 +1,6 @@\n #!/usr/bin/env bash\n-echo hi\n+curl -sf http://localhost:8080/health\n+if [ $? -ne 0 ]; then\n+  echo \"unhealthy\" >&2\n+  exit 1\n+fi\n+echo \"ok $(date)\"\n"},
+			{Role: "tool", Content: "--- a/healthcheck.sh\n+++ b/healthcheck.sh\n@@ -1,2 +1,6 @@\n #!/usr/bin/env bash\n-echo hi\n+curl -sf http://localhost:8080/health\n+if [ $? -ne 0 ]; then\n+  echo \"unhealthy\" >&2\n+  exit 1\n+fi\n+echo \"ok $(date)\"\n"},
 		}
 		m.overlay = OverlayDiff
 

apps/tui-go/internal/app/model.go

@@ -50,8 +50,15 @@ const (
 
 // RouterEntry is one line in a session's conversation transcript.
 type RouterEntry struct {
-	Role    string // user | router | tool
+	Role    string // user | router | tool | narration | narration_llm
 	Content string
+	Metrics *TurnMetrics
+}
+
+// TurnMetrics carries optional latency + token cost for a ROUTER chat turn.
+type TurnMetrics struct {
+	LatencyMs   int64
+	TotalTokens int
 }
 
 // EventEntry is a row in the event stream.

apps/tui-go/internal/app/server.go

@@ -98,7 +98,11 @@ func (m *Model) applyServer(msg protocol.ServerMessage) {
 		if msg.Role == "USER" {
 			role = "user"
 		}
-		m.appendRouter(msg.SessionID, RouterEntry{role, msg.Content})
+		entry := RouterEntry{Role: role, Content: msg.Content}
+		if msg.LatencyMs != nil && msg.TotalTokens != nil {
+			entry.Metrics = &TurnMetrics{LatencyMs: *msg.LatencyMs, TotalTokens: *msg.TotalTokens}
+		}
+		m.appendRouter(msg.SessionID, entry)
 		if s := m.session(msg.SessionID); s != nil {
 			s.LastEventAt = nowMillis()
 		}
@@ -182,7 +186,7 @@ func (m *Model) applyServer(msg protocol.ServerMessage) {
 			s.addEvent(msg.OccurredAt, "ToolCompleted", msg.ToolName)
 		}
 		if msg.Diff != nil && *msg.Diff != "" {
-			m.appendRouter(msg.SessionID, RouterEntry{"tool", *msg.Diff})
+			m.appendRouter(msg.SessionID, RouterEntry{Role: "tool", Content: *msg.Diff})
 		}
 	case protocol.TypeToolAssessed:
 		if s := m.session(msg.SessionID); s != nil {

apps/tui-go/internal/app/view.go

@@ -1,6 +1,7 @@
 package app
 
 import (
+	"fmt"
 	"os"
 	"strings"
 
@@ -352,6 +353,9 @@ func (m Model) routerRows(w, h int) []string {
 			for _, ln := range wrap(e.Content, w) {
 				rows = append(rows, t.span(ln, t.P.Fg))
 			}
+			if s := metricsSuffix(e.Metrics); s != "" {
+				rows = append(rows, t.span(s, t.P.Faint))
+			}
 		case "tool":
 			rows = append(rows, t.span("· tool output ("+itoaLen(e.Content)+" chars) — ^x to view", t.P.Dim))
 		case "narration", "stage":
@@ -570,6 +574,14 @@ func itoa(n int) string {
 
 func itoaLen(s string) string { return itoa(len(s)) }
 
+// metricsSuffix formats the faint latency+token annotation for a ROUTER turn.
+func metricsSuffix(mtr *TurnMetrics) string {
+	if mtr == nil {
+		return ""
+	}
+	return fmt.Sprintf("  %.1fs · %d tokens", float64(mtr.LatencyMs)/1000.0, mtr.TotalTokens)
+}
+
 // wrap splits s into lines no wider than w (rune-based, whitespace-greedy).
 func wrap(s string, w int) []string {
 	if w < 1 {

apps/tui-go/internal/protocol/golden_test.go

@@ -88,6 +88,17 @@ func TestDecodeGoldenServerFrames(t *testing.T) {
 				}
 			},
 		},
+		{
+			name:      "chat.turn carries router metrics",
+			json:      `{"type":"chat.turn","sessionId":"s1","turnId":"t9","role":"ROUTER","content":"hi","latencyMs":3000,"totalTokens":943,"sequence":7,"sessionSequence":2}`,
+			wantType:  TypeChatTurn,
+			eventBear: true,
+			check: func(t *testing.T, m ServerMessage) {
+				if m.LatencyMs == nil || *m.LatencyMs != 3000 || m.TotalTokens == nil || *m.TotalTokens != 943 {
+					t.Fatalf("chat.turn metrics: %+v", m)
+				}
+			},
+		},
 		{
 			name:      "snapshot_complete (non-event control frame)",
 			json:      `{"type":"snapshot_complete"}`,

apps/tui-go/internal/protocol/protocol.go

@@ -87,6 +87,10 @@ type ServerMessage struct {
 	Models  []string `json:"models"`
 	Current string   `json:"current"`
 
+	// chat.turn router metrics (nil when not a ROUTER turn or server omits them)
+	LatencyMs   *int64 `json:"latencyMs"`
+	TotalTokens *int   `json:"totalTokens"`
+
 	// resource.status (nil = unavailable)
 	GpuMemoryUsedMb   *int64 `json:"gpuMemoryUsedMb"`
 	GpuMemoryTotalMb  *int64 `json:"gpuMemoryTotalMb"`