fix(agent/runtimes): show Cursor Composer token usage and billing (#4135)

* fix(agent/runtimes): show Cursor Composer token usage and billing

Attribute Cursor stream-json usage to the configured runtime model when
result events omit `model`, and add Composer/Auto pricing so dashboard
cost estimates resolve for composer-2.5 runs.

Co-authored-by: Cursor <cursoragent@cursor.com>

* fix(views): align Cursor Composer pricing

Co-authored-by: multica-agent <github@multica.ai>

---------

Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Eve <eve@multica-ai.local>
Co-authored-by: multica-agent <github@multica.ai>

packages/views/runtimes/utils.test.ts

@@ -263,6 +263,39 @@ describe("estimateCost", () => {
     ).toBe(0);
   });
 
+  it("prices Cursor Composer rows at the published rates without cache-write spend", () => {
+    const costWithAllTokenTypes = (model: string) =>
+      estimateCost({
+        ...zeroUsage,
+        model,
+        input_tokens: 1_000_000,
+        output_tokens: 1_000_000,
+        cache_read_tokens: 1_000_000,
+        cache_write_tokens: 1_000_000,
+      });
+
+    expect(costWithAllTokenTypes("auto")).toBeCloseTo(1.25 + 6 + 0.25, 5);
+    expect(costWithAllTokenTypes("composer-2.5-fast")).toBeCloseTo(
+      3 + 15 + 0.5,
+      5,
+    );
+    expect(costWithAllTokenTypes("composer-2.5")).toBeCloseTo(0.5 + 2.5 + 0.2, 5);
+    expect(costWithAllTokenTypes("composer-2-fast")).toBeCloseTo(
+      1.5 + 7.5 + 0.35,
+      5,
+    );
+    expect(costWithAllTokenTypes("composer-2")).toBeCloseTo(0.5 + 2.5 + 0.2, 5);
+    expect(costWithAllTokenTypes("composer-1.5")).toBeCloseTo(
+      3.5 + 17.5 + 0.35,
+      5,
+    );
+    expect(costWithAllTokenTypes("composer-1")).toBeCloseTo(
+      1.25 + 10 + 0.125,
+      5,
+    );
+    expect(costWithAllTokenTypes("cursor")).toBeCloseTo(3 + 15 + 0.5, 5);
+  });
+
   // The Chinese-model rates below are spot-checked against the literal
   // numbers on the three official price sheets cited in MODEL_PRICING's
   // header comment. Pinning them in tests is what catches a future edit

packages/views/runtimes/utils.ts

@@ -228,6 +228,24 @@ const MODEL_PRICING: Record<
   "glm-4.5-air":        { input: 0.2,  output: 1.1,  cacheRead: 0.03,   cacheWrite: 0.2 },
   "glm-4.5-airx":       { input: 1.1,  output: 4.5,  cacheRead: 0.22,   cacheWrite: 1.1 },
   "glm-4.5-flash":      { input: 0,    output: 0,    cacheRead: 0,      cacheWrite: 0 },
+
+  // -- Cursor Composer / Auto (cursor.com/docs/models-and-pricing,
+  //    cursor.com/docs/models/cursor-composer-2,
+  //    cursor.com/docs/models/cursor-composer-2-5).
+  //    Cursor result events often omit `model`, so the daemon falls back to
+  //    the configured runtime model or the legacy key `cursor`.
+  //    Cursor does not publish a cache-write rate for these rows; keep it at
+  //    0 so reported cache_write_tokens don't invent spend from input pricing.
+  "auto":               { input: 1.25, output: 6,    cacheRead: 0.25,   cacheWrite: 0 },
+  "composer-2.5-fast":  { input: 3,    output: 15,   cacheRead: 0.5,    cacheWrite: 0 },
+  "composer-2.5":       { input: 0.5,  output: 2.5,  cacheRead: 0.2,    cacheWrite: 0 },
+  "composer-2-fast":    { input: 1.5,  output: 7.5,  cacheRead: 0.35,   cacheWrite: 0 },
+  "composer-2":         { input: 0.5,  output: 2.5,  cacheRead: 0.2,    cacheWrite: 0 },
+  "composer-1.5":       { input: 3.5,  output: 17.5, cacheRead: 0.35,   cacheWrite: 0 },
+  "composer-1":         { input: 1.25, output: 10,   cacheRead: 0.125,  cacheWrite: 0 },
+  // Legacy fallback bucket when neither the result event nor the runtime
+  // model is known — price at the current Composer 2.5 Fast default.
+  "cursor":             { input: 3,    output: 15,   cacheRead: 0.5,    cacheWrite: 0 },
 };
 
 // Resolve a model string to its pricing tier. Exact match, with four

server/pkg/agent/cursor.go

@@ -74,6 +74,7 @@ func (b *cursorBackend) Execute(ctx context.Context, prompt string, opts ExecOpt
 		}()
 
 		startTime := time.Now()
+		configuredModel := strings.TrimSpace(opts.Model)
 		var output strings.Builder
 		var sessionID string
 		finalStatus := "completed"
@@ -149,7 +150,7 @@ func (b *cursorBackend) Execute(ctx context.Context, prompt string, opts ExecOpt
 				if evt.ResultText != "" && output.Len() == 0 {
 					output.WriteString(evt.ResultText)
 				}
-				b.accumulateResultUsage(resultUsage, &evt)
+				b.accumulateResultUsage(resultUsage, &evt, configuredModel)
 				if evt.hasResultUsage() {
 					hasResultUsage = true
 				}
@@ -179,10 +180,7 @@ func (b *cursorBackend) Execute(ctx context.Context, prompt string, opts ExecOpt
 				if evt.Part != nil {
 					var part cursorStepFinishPart
 					_ = json.Unmarshal(evt.Part, &part)
-					model := evt.Model
-					if model == "" {
-						model = "cursor"
-					}
+					model := cursorUsageModel(evt.Model, configuredModel)
 					u := stepUsage[model]
 					u.InputTokens += int64(part.Tokens.Input)
 					u.OutputTokens += int64(part.Tokens.Output)
@@ -267,11 +265,18 @@ func (b *cursorBackend) handleCursorAssistant(evt *cursorStreamEvent, ch chan<-
 	}
 }
 
-func (b *cursorBackend) accumulateResultUsage(usage map[string]TokenUsage, evt *cursorStreamEvent) {
-	model := evt.Model
-	if model == "" {
-		model = "cursor"
+func cursorUsageModel(evtModel, configuredModel string) string {
+	if model := strings.TrimSpace(evtModel); model != "" {
+		return model
 	}
+	if model := strings.TrimSpace(configuredModel); model != "" {
+		return model
+	}
+	return "cursor"
+}
+
+func (b *cursorBackend) accumulateResultUsage(usage map[string]TokenUsage, evt *cursorStreamEvent, configuredModel string) {
+	model := cursorUsageModel(evt.Model, configuredModel)
 	u := usage[model]
 
 	// Cursor agent has emitted token usage in multiple shapes: top-level

server/pkg/agent/cursor_test.go

@@ -251,6 +251,49 @@ func TestCursorErrorText(t *testing.T) {
 	}
 }
 
+func TestCursorUsageModelFallback(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name             string
+		evtModel         string
+		configuredModel  string
+		want             string
+	}{
+		{"event model wins", "gpt-5.3-codex", "composer-2.5", "gpt-5.3-codex"},
+		{"configured model fallback", "", "composer-2.5", "composer-2.5"},
+		{"default cursor", "", "", "cursor"},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			got := cursorUsageModel(tc.evtModel, tc.configuredModel)
+			if got != tc.want {
+				t.Fatalf("cursorUsageModel(%q, %q) = %q, want %q", tc.evtModel, tc.configuredModel, got, tc.want)
+			}
+		})
+	}
+}
+
+func TestCursorAccumulateResultUsageUsesConfiguredModel(t *testing.T) {
+	t.Parallel()
+
+	b := &cursorBackend{cfg: Config{Logger: slog.Default()}}
+	usage := make(map[string]TokenUsage)
+	evt := &cursorStreamEvent{
+		InputTokens:  400,
+		OutputTokens: 200,
+	}
+	b.accumulateResultUsage(usage, evt, "composer-2.5")
+	u := usage["composer-2.5"]
+	if u.InputTokens != 400 || u.OutputTokens != 200 {
+		t.Fatalf("unexpected usage: %+v", u)
+	}
+	if _, ok := usage["cursor"]; ok {
+		t.Fatalf("expected configured model key, got cursor fallback: %+v", usage)
+	}
+}
+
 func TestCursorAccumulateResultUsage(t *testing.T) {
 	t.Parallel()
 
@@ -269,7 +312,7 @@ func TestCursorAccumulateResultUsage(t *testing.T) {
 				CacheWriteInputTokens: 25,
 			},
 		}
-		b.accumulateResultUsage(usage, evt)
+		b.accumulateResultUsage(usage, evt, "")
 		u := usage["gpt-5.3"]
 		if u.InputTokens != 200 || u.OutputTokens != 100 || u.CacheReadTokens != 50 || u.CacheWriteTokens != 25 {
 			t.Fatalf("unexpected usage: %+v", u)
@@ -288,7 +331,7 @@ func TestCursorAccumulateResultUsage(t *testing.T) {
 			CacheReadTokens:  75,
 			CacheWriteTokens: 25,
 		}
-		b.accumulateResultUsage(usage, evt)
+		b.accumulateResultUsage(usage, evt, "")
 		u := usage["gpt-5.3"]
 		if u.InputTokens != 300 || u.OutputTokens != 150 || u.CacheReadTokens != 75 || u.CacheWriteTokens != 25 {
 			t.Fatalf("unexpected usage: %+v (want input=300 output=150 cache_read=75 cache_write=25)", u)
@@ -309,7 +352,7 @@ func TestCursorAccumulateResultUsage(t *testing.T) {
 				CacheReadInputTokens: 777,
 			},
 		}
-		b.accumulateResultUsage(usage, evt)
+		b.accumulateResultUsage(usage, evt, "")
 		u := usage["gpt-5.3"]
 		if u.InputTokens != 300 || u.OutputTokens != 150 || u.CacheReadTokens != 0 {
 			t.Fatalf("unexpected usage: %+v (want input=300 output=150 cache=0)", u)
@@ -322,7 +365,7 @@ func TestCursorAccumulateResultUsage(t *testing.T) {
 		evt := &cursorStreamEvent{
 			Model: "gpt-5.3",
 		}
-		b.accumulateResultUsage(usage, evt)
+		b.accumulateResultUsage(usage, evt, "")
 		if _, ok := usage["gpt-5.3"]; ok {
 			t.Fatalf("expected no entry, got %+v", usage["gpt-5.3"])
 		}
@@ -335,7 +378,7 @@ func TestCursorAccumulateResultUsage(t *testing.T) {
 			InputTokens:  50,
 			OutputTokens: 25,
 		}
-		b.accumulateResultUsage(usage, evt)
+		b.accumulateResultUsage(usage, evt, "")
 		u := usage["cursor"]
 		if u.InputTokens != 50 || u.OutputTokens != 25 {
 			t.Fatalf("unexpected usage: %+v (want input=50 output=25)", u)
@@ -502,7 +545,7 @@ func TestCursorUsageNoDoubleCount(t *testing.T) {
 
 				switch evt.Type {
 				case "result":
-					b.accumulateResultUsage(resultUsage, &evt)
+					b.accumulateResultUsage(resultUsage, &evt, "")
 					if evt.hasResultUsage() {
 						hasResultUsage = true
 					}
@@ -580,7 +623,7 @@ func TestCursorStreamEventUnmarshalTopLevelCamelCase(t *testing.T) {
 	// Verify accumulateResultUsage processes the new shape.
 	b := &cursorBackend{cfg: Config{Logger: slog.Default()}}
 	usage := make(map[string]TokenUsage)
-	b.accumulateResultUsage(usage, &evt)
+	b.accumulateResultUsage(usage, &evt, "")
 	u := usage["gpt-5.3"]
 	if u.InputTokens != 1500 || u.OutputTokens != 300 || u.CacheReadTokens != 75 || u.CacheWriteTokens != 25 {
 		t.Fatalf("accumulated usage = %+v, want input=1500 output=300 cache_read=75 cache_write=25", u)
@@ -607,7 +650,7 @@ func TestCursorStreamEventUnmarshalNestedCamelCase(t *testing.T) {
 
 	b := &cursorBackend{cfg: Config{Logger: slog.Default()}}
 	usage := make(map[string]TokenUsage)
-	b.accumulateResultUsage(usage, &evt)
+	b.accumulateResultUsage(usage, &evt, "")
 	u := usage["cursor"]
 	if u.InputTokens != 26640 || u.OutputTokens != 40 || u.CacheReadTokens != 467 || u.CacheWriteTokens != 12 {
 		t.Fatalf("accumulated usage = %+v, want input=26640 output=40 cache_read=467 cache_write=12", u)
@@ -637,7 +680,7 @@ func TestCursorStreamEventUnmarshalLegacyUsage(t *testing.T) {
 
 	b := &cursorBackend{cfg: Config{Logger: slog.Default()}}
 	usage := make(map[string]TokenUsage)
-	b.accumulateResultUsage(usage, &evt)
+	b.accumulateResultUsage(usage, &evt, "")
 	u := usage["gpt-5"]
 	if u.InputTokens != 800 || u.OutputTokens != 400 || u.CacheReadTokens != 200 || u.CacheWriteTokens != 100 {
 		t.Fatalf("accumulated usage = %+v, want input=800 output=400 cache_read=200 cache_write=100", u)