show cached input hint with input/output tokens (codex style)

src/lib/entries.ts

@@ -69,13 +69,17 @@ export interface QuotaToastError {
 export interface SessionTokenModel {
   modelID: string;
   input: number;
+  cachedInput?: number;
+  totalInput?: number;
   output: number;
 }
 
 /** Session tokens data for toast display. */
 export interface SessionTokensData {
   models: SessionTokenModel[];
   totalInput: number;
+  totalCachedInput?: number;
+  totalCombinedInput?: number;
   totalOutput: number;
 }
 

src/lib/quota-stats.ts

@@ -821,13 +821,17 @@ export async function aggregateUsage(params: {
 export type SessionTokenRow = {
   modelID: string;
   input: number;
+  cachedInput: number;
+  totalInput: number;
   output: number;
 };
 
 export type SessionTokenSummary = {
   sessionID: string;
   models: SessionTokenRow[];
   totalInput: number;
+  totalCachedInput: number;
+  totalCombinedInput: number;
   totalOutput: number;
 };
 
@@ -839,43 +843,59 @@ export async function getSessionTokenSummary(
 
   if (sessionMessages.length === 0) return null;
 
-  const byModel = new Map<string, { input: number; output: number }>();
+  const byModel = new Map<string, { input: number; cachedInput: number; totalInput: number; output: number }>();
   let totalInput = 0;
+  let totalCachedInput = 0;
+  let totalCombinedInput = 0;
   let totalOutput = 0;
 
   for (const msg of sessionMessages) {
     const tokens = msg.tokens;
     if (!tokens) continue;
 
     const input = typeof tokens.input === "number" ? tokens.input : 0;
+    const cachedInput = typeof tokens.cache?.read === "number" ? tokens.cache.read : 0;
+    const totalInputForMessage = input + cachedInput;
     const output = typeof tokens.output === "number" ? tokens.output : 0;
 
     // Skip if both are 0
-    if (input === 0 && output === 0) continue;
+    if (totalInputForMessage === 0 && output === 0) continue;
 
     totalInput += input;
+    totalCachedInput += cachedInput;
+    totalCombinedInput += totalInputForMessage;
     totalOutput += output;
 
     const modelID = msg.modelID ?? "unknown";
     const existing = byModel.get(modelID);
     if (existing) {
       existing.input += input;
+      existing.cachedInput += cachedInput;
+      existing.totalInput += totalInputForMessage;
       existing.output += output;
     } else {
-      byModel.set(modelID, { input, output });
+      byModel.set(modelID, { input, cachedInput, totalInput: totalInputForMessage, output });
     }
   }
 
   // Sort by total tokens descending
   const models = Array.from(byModel.entries())
-    .map(([modelID, t]) => ({ modelID, input: t.input, output: t.output }))
-    .filter((m) => m.input > 0 || m.output > 0)
-    .sort((a, b) => b.input + b.output - (a.input + a.output));
+    .map(([modelID, t]) => ({
+      modelID,
+      input: t.input,
+      cachedInput: t.cachedInput,
+      totalInput: t.totalInput,
+      output: t.output,
+    }))
+    .filter((m) => m.totalInput > 0 || m.output > 0)
+    .sort((a, b) => b.totalInput + b.output - (a.totalInput + a.output));
 
   return {
     sessionID,
     models,
     totalInput,
+    totalCachedInput,
+    totalCombinedInput,
     totalOutput,
   };
 }

src/lib/session-tokens-format.ts

@@ -15,6 +15,16 @@ export type SessionTokenSectionModel = {
   lines: string[];
 };
 
+function hasCachedInput(model: SessionTokensData["models"][number]): boolean {
+  return (model.cachedInput ?? 0) > 0;
+}
+
+function hasAnyCachedInput(sessionTokens: SessionTokensData): boolean {
+  return (
+    (sessionTokens.totalCachedInput ?? 0) > 0 || sessionTokens.models.some((model) => hasCachedInput(model))
+  );
+}
+
 function normalizeMaxWidth(maxWidth?: number): number | undefined {
   if (typeof maxWidth !== "number" || !Number.isFinite(maxWidth)) return undefined;
   return Math.max(1, Math.trunc(maxWidth));
@@ -27,11 +37,20 @@ function clampRenderedLine(line: string, maxWidth?: number): string {
 
 function buildWideSessionTokenSectionModel(sessionTokens: SessionTokensData): SessionTokenSectionModel {
   const lines: string[] = [];
+  const showCached = hasAnyCachedInput(sessionTokens);
   for (const model of sessionTokens.models) {
     const shortName = shortenModelName(model.modelID, 20);
     const inStr = formatTokenCount(model.input);
+    const cachedStr = formatTokenCount(model.cachedInput ?? 0);
+    const totalInStr = formatTokenCount(model.totalInput ?? model.input + (model.cachedInput ?? 0));
     const outStr = formatTokenCount(model.output);
-    lines.push(`  ${padRight(shortName, 20)}  ${padLeft(inStr, 6)} in  ${padLeft(outStr, 6)} out`);
+    if (showCached) {
+      lines.push(
+        `  ${padRight(shortName, 20)}  ${padLeft(inStr, 6)} new  ${padLeft(cachedStr, 6)} cache  ${padLeft(totalInStr, 6)} in  ${padLeft(outStr, 6)} out`,
+      );
+    } else {
+      lines.push(`  ${padRight(shortName, 20)}  ${padLeft(inStr, 6)} in  ${padLeft(outStr, 6)} out`);
+    }
   }
 
   return {
@@ -46,14 +65,19 @@ function buildCompactSessionTokenSectionModel(
 ): SessionTokenSectionModel {
   const width = Math.max(1, Math.trunc(maxWidth));
   const lines: string[] = [];
+  const showCached = hasAnyCachedInput(sessionTokens);
 
   for (const model of sessionTokens.models) {
     const modelIndent = width > 2 ? "  " : "";
     const modelLineWidth = Math.max(1, width - modelIndent.length);
     const detailIndent = width > 4 ? "    " : width > 2 ? "  " : "";
     const inStr = formatTokenCount(model.input);
+    const cachedStr = formatTokenCount(model.cachedInput ?? 0);
+    const totalInStr = formatTokenCount(model.totalInput ?? model.input + (model.cachedInput ?? 0));
     const outStr = formatTokenCount(model.output);
-    const compactCounts = `${inStr} in  ${outStr} out`;
+    const compactCounts = showCached
+      ? `${inStr} new  ${cachedStr} cache  ${totalInStr} in  ${outStr} out`
+      : `${inStr} in  ${outStr} out`;
 
     lines.push(`${modelIndent}${shortenModelName(model.modelID, modelLineWidth)}`.slice(0, width));
 
@@ -62,8 +86,15 @@ function buildCompactSessionTokenSectionModel(
       continue;
     }
 
-    lines.push(`${detailIndent}${inStr} in`.slice(0, width));
-    lines.push(`${detailIndent}${outStr} out`.slice(0, width));
+    if (showCached) {
+      lines.push(`${detailIndent}${inStr} new`.slice(0, width));
+      lines.push(`${detailIndent}${cachedStr} cache`.slice(0, width));
+      lines.push(`${detailIndent}${totalInStr} in`.slice(0, width));
+      lines.push(`${detailIndent}${outStr} out`.slice(0, width));
+    } else {
+      lines.push(`${detailIndent}${inStr} in`.slice(0, width));
+      lines.push(`${detailIndent}${outStr} out`.slice(0, width));
+    }
   }
 
   return {
@@ -76,7 +107,11 @@ function buildSidebarSessionTokenSummaryModel(
   sessionTokens: SessionTokensData,
   options?: { maxWidth?: number },
 ): SessionTokenSectionModel {
-  const summaryLine = `  ${formatTokenCount(sessionTokens.totalInput)} in  ${formatTokenCount(sessionTokens.totalOutput)} out`;
+  const totalCached = sessionTokens.totalCachedInput ?? 0;
+  const totalCombined = sessionTokens.totalCombinedInput ?? sessionTokens.totalInput + totalCached;
+  const summaryLine = totalCached > 0
+    ? `  ${formatTokenCount(sessionTokens.totalInput)} new  ${formatTokenCount(totalCached)} cache  ${formatTokenCount(totalCombined)} in  ${formatTokenCount(sessionTokens.totalOutput)} out`
+    : `  ${formatTokenCount(sessionTokens.totalInput)} in  ${formatTokenCount(sessionTokens.totalOutput)} out`;
   return {
     heading: clampRenderedLine(SESSION_TOKEN_SECTION_HEADING, options?.maxWidth),
     lines: [clampRenderedLine(summaryLine, options?.maxWidth)],

src/lib/session-tokens.ts

@@ -35,6 +35,8 @@ export async function fetchSessionTokensForDisplay(params: {
         sessionTokens: {
           models: summary.models,
           totalInput: summary.totalInput,
+          totalCachedInput: summary.totalCachedInput,
+          totalCombinedInput: summary.totalCombinedInput,
           totalOutput: summary.totalOutput,
         },
       };

src/lib/tui-compact-format.ts

@@ -168,9 +168,23 @@ function formatCompactSessionTokensSegment(data: QuotaRenderData): string | null
   const hasTokenData =
     sessionTokens.models.length > 0 ||
     sessionTokens.totalInput > 0 ||
+    (sessionTokens.totalCachedInput ?? 0) > 0 ||
     sessionTokens.totalOutput > 0;
   if (!hasTokenData) return null;
 
+  const totalCached = sessionTokens.totalCachedInput ?? 0;
+  const totalCombined = sessionTokens.totalCombinedInput ?? sessionTokens.totalInput + totalCached;
+
+  if (totalCached > 0) {
+    return compactText(
+      `tok ${formatCompactTokenCount(sessionTokens.totalInput)} new / ${formatCompactTokenCount(
+        totalCached,
+      )} cache / ${formatCompactTokenCount(totalCombined)} in / ${formatCompactTokenCount(
+        sessionTokens.totalOutput,
+      )} out`,
+    );
+  }
+
   return compactText(
     `tok ${formatCompactTokenCount(sessionTokens.totalInput)} in / ${formatCompactTokenCount(
       sessionTokens.totalOutput,

tests/format.test.ts

@@ -469,6 +469,35 @@ describe("formatQuotaRows", () => {
     expect(out).not.toContain("openai/gpt-5.4-mini");
   });
 
+  it("renders single-window session tokens with new and cached input totals when available", () => {
+    const out = formatQuotaRows({
+      version: "1.0.0",
+      style: "singleWindow",
+      layout: { maxWidth: 80, narrowAt: 32, tinyAt: 20 },
+      entries: [],
+      sessionTokens: {
+        totalInput: 372,
+        totalCachedInput: 120,
+        totalCombinedInput: 492,
+        totalOutput: 41,
+        models: [
+          {
+            modelID: "openai/gpt-5.4-mini",
+            input: 372,
+            cachedInput: 120,
+            totalInput: 492,
+            output: 41,
+          },
+        ],
+      },
+    });
+
+    expect(out.split("\n")).toEqual([
+      SESSION_TOKEN_SECTION_HEADING,
+      "  372 new  120 cache  492 in  41 out",
+    ]);
+  });
+
   it("renders all-window session tokens with detailed per-model rows", () => {
     const out = formatQuotaRows({
       version: "1.0.0",
@@ -489,6 +518,35 @@ describe("formatQuotaRows", () => {
     ]);
   });
 
+  it("renders all-window session tokens with separate new and cached input when available", () => {
+    const out = formatQuotaRows({
+      version: "1.0.0",
+      style: "allWindows",
+      layout: { maxWidth: 80, narrowAt: 32, tinyAt: 20 },
+      entries: [],
+      sessionTokens: {
+        totalInput: 372,
+        totalCachedInput: 120,
+        totalCombinedInput: 492,
+        totalOutput: 41,
+        models: [
+          {
+            modelID: "openai/gpt-5.4-mini",
+            input: 372,
+            cachedInput: 120,
+            totalInput: 492,
+            output: 41,
+          },
+        ],
+      },
+    });
+
+    expect(out.split("\n")).toEqual([
+      SESSION_TOKEN_SECTION_HEADING,
+      "  openai/gpt-5.4-mini      372 new     120 cache     492 in      41 out",
+    ]);
+  });
+
   it("keeps legacy style aliases working for direct formatter calls", () => {
     const aliasOutput = formatQuotaRows({
       version: "1.0.0",

tests/quota-command-format.test.ts

@@ -52,10 +52,12 @@ describe("formatQuotaCommand", () => {
       errors: [{ label: "Z.ai", message: "Authentication expired" }],
       sessionTokens: {
         models: [
-          { modelID: "openai/gpt-5", input: 1234, output: 567 },
+          { modelID: "openai/gpt-5", input: 1234, cachedInput: 456, totalInput: 1690, output: 567 },
           { modelID: "github-copilot/claude-sonnet-4.5", input: 987, output: 654 },
         ],
         totalInput: 2221,
+        totalCachedInput: 456,
+        totalCombinedInput: 2677,
         totalOutput: 1221,
       },
     });
@@ -78,8 +80,8 @@ describe("formatQuotaCommand", () => {
         Claude:          ████████████░░░░░░  67% left (resets in 3h)
 
       Session input/output tokens
-        openai/gpt-5            1.2K in     567 out
-        github-copilot/clau…     987 in     654 out
+        openai/gpt-5            1.2K new     456 cache    1.7K in     567 out
+        github-copilot/clau…     987 new       0 cache     987 in     654 out
 
       Z.ai: Authentication expired"
     `);

tests/session-tokens-format.test.ts

@@ -53,6 +53,29 @@ describe("renderSessionTokensLines", () => {
       lines: ["  openai/gpt-5            1.2K in     567 out"],
     });
   });
+
+  it("renders separate new and cached input when cache data is present", () => {
+    const lines = renderSessionTokensLines({
+      models: [
+        {
+          modelID: "openai/gpt-5",
+          input: 1234,
+          cachedInput: 456,
+          totalInput: 1690,
+          output: 567,
+        },
+      ],
+      totalInput: 1234,
+      totalCachedInput: 456,
+      totalCombinedInput: 1690,
+      totalOutput: 567,
+    });
+
+    expect(lines).toEqual([
+      SESSION_TOKEN_SECTION_HEADING,
+      "  openai/gpt-5            1.2K new     456 cache    1.7K in     567 out",
+    ]);
+  });
 });
 
 describe("renderSidebarSessionTokenSummaryLines", () => {
@@ -69,4 +92,27 @@ describe("renderSidebarSessionTokenSummaryLines", () => {
     expect(lines).toEqual([SESSION_TOKEN_SECTION_HEADING.slice(0, 36), "  372 in  41 out"]);
     expect(lines.every((line) => line.length <= 36)).toBe(true);
   });
+
+  it("renders aggregate new and cached input in the sidebar summary when cache data exists", () => {
+    const lines = renderSidebarSessionTokenSummaryLines(
+      {
+        models: [
+          {
+            modelID: "openai/gpt-5.4-mini",
+            input: 372,
+            cachedInput: 120,
+            totalInput: 492,
+            output: 41,
+          },
+        ],
+        totalInput: 372,
+        totalCachedInput: 120,
+        totalCombinedInput: 492,
+        totalOutput: 41,
+      },
+      { maxWidth: 80 },
+    );
+
+    expect(lines).toEqual([SESSION_TOKEN_SECTION_HEADING, "  372 new  120 cache  492 in  41 out"]);
+  });
 });

tests/tui-compact-format.test.ts

@@ -123,14 +123,16 @@ describe("buildCompactQuotaStatusLine", () => {
         ],
         errors: [],
         sessionTokens: {
-          models: [{ modelID: "openai/gpt-5", input: 12_400, output: 3_100 }],
+          models: [{ modelID: "openai/gpt-5", input: 12_400, cachedInput: 5_600, totalInput: 18_000, output: 3_100 }],
           totalInput: 12_400,
+          totalCachedInput: 5_600,
+          totalCombinedInput: 18_000,
           totalOutput: 3_100,
         },
       },
     });
 
-    expect(line).toBe("Copilot 82% | Cursor API - $2.40 | tok 12.4K in / 3.1K out");
+    expect(line).toBe("Copilot 82% | Cursor API - $2.40 | tok 12.4K new / 5.6K cache / 18K in / 3.1K out");
   });
 
   it("summarizes errors as issue counts when quota segments exist and the count fits", () => {