tokenmaxx 0.0.3 → 0.0.5

package/dist/index.js

@@ -14800,6 +14800,7 @@ var init_domain = __esm(() => {
   TokenEventSchema = exports_external.object({
     accountId: exports_external.uuid().nullable(),
     at: exports_external.number().int().nonnegative(),
+    cacheReadTokens: exports_external.number().int().nonnegative(),
     inputTokens: exports_external.number().int().nonnegative(),
     model: exports_external.string().min(1).nullable(),
     outputTokens: exports_external.number().int().nonnegative(),
@@ -14814,6 +14815,7 @@ var init_domain = __esm(() => {
     key: exports_external.string(),
     peakPerHour: exports_external.number().nonnegative(),
     topModel: exports_external.string().nullable(),
+    totalCached: exports_external.number().nonnegative(),
     totalInput: exports_external.number().nonnegative(),
     totalOutput: exports_external.number().nonnegative(),
     totalTokens: exports_external.number().nonnegative()
@@ -15087,21 +15089,21 @@ function priceFor(model) {
   const key = (model ?? "").toLowerCase();
   return PRICES.find((entry) => key.includes(entry.match))?.price ?? DEFAULT_PRICE;
 }
-function costUsd(model, inputTokens, outputTokens) {
+function costUsd(model, inputTokens, outputTokens, cacheReadTokens = 0) {
   const price = priceFor(model);
-  return inputTokens / 1e6 * price.inputPerMTok + outputTokens / 1e6 * price.outputPerMTok;
+  return inputTokens / 1e6 * price.inputPerMTok + outputTokens / 1e6 * price.outputPerMTok + cacheReadTokens / 1e6 * price.cacheReadPerMTok;
 }
 var PRICES, DEFAULT_PRICE;
 var init_pricing = __esm(() => {
   PRICES = [
-    { match: "claude-opus", price: { inputPerMTok: 15, outputPerMTok: 75 } },
-    { match: "claude-sonnet", price: { inputPerMTok: 3, outputPerMTok: 15 } },
-    { match: "claude-haiku", price: { inputPerMTok: 0.8, outputPerMTok: 4 } },
-    { match: "claude-fable", price: { inputPerMTok: 1, outputPerMTok: 5 } },
-    { match: "gpt-5", price: { inputPerMTok: 1.25, outputPerMTok: 10 } },
-    { match: "o4", price: { inputPerMTok: 1.1, outputPerMTok: 4.4 } }
+    { match: "claude-opus", price: { cacheReadPerMTok: 1.5, inputPerMTok: 15, outputPerMTok: 75 } },
+    { match: "claude-sonnet", price: { cacheReadPerMTok: 0.3, inputPerMTok: 3, outputPerMTok: 15 } },
+    { match: "claude-haiku", price: { cacheReadPerMTok: 0.08, inputPerMTok: 0.8, outputPerMTok: 4 } },
+    { match: "claude-fable", price: { cacheReadPerMTok: 0.1, inputPerMTok: 1, outputPerMTok: 5 } },
+    { match: "gpt-5", price: { cacheReadPerMTok: 0.125, inputPerMTok: 1.25, outputPerMTok: 10 } },
+    { match: "o4", price: { cacheReadPerMTok: 0.28, inputPerMTok: 1.1, outputPerMTok: 4.4 } }
   ];
-  DEFAULT_PRICE = { inputPerMTok: 3, outputPerMTok: 15 };
+  DEFAULT_PRICE = { cacheReadPerMTok: 0.3, inputPerMTok: 3, outputPerMTok: 15 };
 });
 
 // src/tui/format.ts
@@ -15141,32 +15143,24 @@ function meter(usedPercent, width = 14) {
 function percentLabel(usedPercent) {
   return usedPercent === null ? "  ?%" : `${Math.round(clamp(usedPercent))}%`.padStart(4);
 }
-function brailleLine(columns, width, height, max = 100) {
+function brailleArea(columns, width, height, max = 100) {
   const dotRows = height * 4;
   const dotCols = width * 2;
   const scale = max <= 0 ? 1 : max;
   const grid = Array.from({ length: dotCols }, () => new Array(dotRows).fill(false));
-  const toY = (value) => Math.max(0, Math.min(dotRows - 1, Math.round(value / scale * (dotRows - 1))));
-  let previousY = -1;
   for (let x = 0;x < dotCols; x += 1) {
     const value = columns[x];
-    if (value === null || value === undefined) {
-      previousY = -1;
+    if (value === null || value === undefined || value <= 0) {
       continue;
     }
-    const y = toY(value);
+    const top = Math.max(1, Math.min(dotRows, Math.round(value / scale * dotRows)));
     const column = grid[x];
     if (column === undefined) {
       continue;
     }
-    if (previousY >= 0) {
-      for (let fill = Math.min(previousY, y);fill <= Math.max(previousY, y); fill += 1) {
-        column[fill] = true;
-      }
-    } else {
+    for (let y = 0;y < top; y += 1) {
       column[y] = true;
     }
-    previousY = y;
   }
   const rows = [];
   for (let charRow = 0;charRow < height; charRow += 1) {
@@ -15366,7 +15360,8 @@ function buildTokens(scale) {
     const target = Math.round(150000 * hours * 0.5 * scale);
     const buckets = raw.map((value) => Math.round(value / rawSum * target));
     const totalTokens = buckets.reduce((sum, value) => sum + value, 0);
-    const totalInput = Math.round(totalTokens * 0.7);
+    const totalCached = Math.round(totalTokens * 0.62);
+    const totalInput = Math.round(totalTokens * 0.28);
     const codexTokens = Math.round(totalTokens * 0.55);
     const claudeTokens = totalTokens - codexTokens;
     const codexCost = costUsd("gpt-5.6-sol", Math.round(codexTokens * 0.7), Math.round(codexTokens * 0.3));
@@ -15384,8 +15379,9 @@ function buildTokens(scale) {
       key: timeframe.key,
       peakPerHour: Math.round(peakBucket * (3600000 / bucketMs)),
       topModel: totalTokens === 0 ? null : "claude-opus-4-8",
+      totalCached,
       totalInput,
-      totalOutput: totalTokens - totalInput,
+      totalOutput: totalTokens - totalInput - totalCached,
       totalTokens
     };
   });
@@ -15808,28 +15804,43 @@ function timeframeBar(ctx, timeframe) {
 function throughputCard(ctx, tokens, timeframe, height, width) {
   const body = [];
   if (tokens === undefined || tokens.totalTokens === 0) {
-    for (let index = 0;index < Math.max(1, height - 1); index += 1) {
+    for (let index = 0;index < Math.max(1, Math.floor(height / 2)); index += 1) {
       body.push(Box({ flexDirection: "row" }, Text({ content: " ", fg: rgb(ctx.theme.bg) })));
     }
-    body.push(Box({ flexDirection: "row" }, Text({ content: "  no token usage yet \u2014 run ", fg: rgb(ctx.theme.dim) }), Text({ content: "codex", fg: rgb(ctx.theme.fg) }), Text({ content: " or ", fg: rgb(ctx.theme.dim) }), Text({ content: "claude", fg: rgb(ctx.theme.fg) }), Text({ content: " and it fills in", fg: rgb(ctx.theme.dim) })));
+    body.push(Box({ flexDirection: "row", justifyContent: "center", width: "100%" }, Text({ content: "no token usage yet \u2014 run ", fg: rgb(ctx.theme.dim) }), Text({ content: "codex", fg: rgb(ctx.theme.fg) }), Text({ content: " or ", fg: rgb(ctx.theme.dim) }), Text({ content: "claude", fg: rgb(ctx.theme.fg) }), Text({ content: " and it fills in live", fg: rgb(ctx.theme.dim) })));
   } else {
-    const columns = throughputColumns(tokens.buckets, width * 2);
+    const axisTop = `${compactNumber(tokens.peakPerHour)}/h`;
+    const gutter = Math.max(6, axisTop.length);
+    const chartWidth = Math.max(16, width - gutter - 1);
+    const columns = throughputColumns(tokens.buckets, chartWidth * 2);
     const peak = Math.max(...columns, 1);
-    const chart = brailleLine(columns, width, height, peak);
-    const axisTop = `${compactNumber(tokens.peakPerHour)}/h`.padStart(6);
+    const chart = brailleArea(columns, chartWidth, height, peak);
     chart.forEach((line, index) => {
-      const axis = index === 0 ? axisTop : index === chart.length - 1 ? "     0" : "      ";
-      body.push(Box({ flexDirection: "row" }, Text({ content: `${axis} `, fg: rgb(ctx.theme.faint) }), Text({ content: line, fg: rgb(ctx.theme.accent) })));
+      const label = index === 0 ? axisTop : index === chart.length - 1 ? "0" : "";
+      body.push(Box({ flexDirection: "row" }, Text({ content: `${label.padStart(gutter)} `, fg: rgb(ctx.theme.faint) }), Text({ content: line, fg: rgb(ctx.theme.accent) })));
     });
-    body.push(Box({ flexDirection: "row" }, Text({ content: "       ", fg: rgb(ctx.theme.bg) }), Text({
-      content: `${timeframe.label} ago`.padEnd(Math.max(0, width - 3)),
+    body.push(Box({ flexDirection: "row" }, Text({ content: " ".repeat(gutter + 1), fg: rgb(ctx.theme.bg) }), Text({
+      content: `${timeframe.label} ago`.padEnd(Math.max(0, chartWidth - 3)),
       fg: rgb(ctx.theme.faint)
     }), Text({ content: "now", fg: rgb(ctx.theme.faint) })));
+    const spanHours = tokens.bucketMs * tokens.buckets.length / 3600000;
+    const averagePerHour = spanHours > 0 ? tokens.totalTokens / spanHours : 0;
+    const recent = tokens.buckets.slice(-2);
+    const nowPerHour = Math.max(...recent, 0) * (3600000 / tokens.bucketMs);
+    const cachedShare = tokens.totalTokens > 0 ? Math.round(tokens.totalCached / tokens.totalTokens * 100) : 0;
     body.push(Box({ flexDirection: "row", paddingLeft: 1 }, Text({ content: "\u03A3 ", fg: rgb(ctx.theme.dim) }), Text({
       attributes: 1,
       content: `${compactNumber(tokens.totalTokens)} tokens`,
       fg: rgb(ctx.theme.fg)
-    }), Text({ content: "   \u2248 ", fg: rgb(ctx.theme.dim) }), Text({ attributes: 1, content: compactUsd(tokens.costUsd), fg: rgb(ctx.theme.good) }), Text({ content: "   peak ", fg: rgb(ctx.theme.dim) }), Text({ content: `${compactNumber(tokens.peakPerHour)}/h`, fg: rgb(ctx.theme.accent) }), Text({
+    }), Text({ content: "   \u2248 ", fg: rgb(ctx.theme.dim) }), Text({ attributes: 1, content: compactUsd(tokens.costUsd), fg: rgb(ctx.theme.good) }), Text({ content: " API value", fg: rgb(ctx.theme.dim) }), Text({
+      content: cachedShare > 0 ? `   ${cachedShare}% cached` : "",
+      fg: rgb(ctx.theme.dim)
+    })));
+    body.push(Box({ flexDirection: "row", paddingLeft: 1 }, Text({ content: "now ", fg: rgb(ctx.theme.dim) }), Text({
+      attributes: 1,
+      content: `${compactNumber(nowPerHour)}/h`,
+      fg: rgb(nowPerHour > 0 ? ctx.theme.accent : ctx.theme.faint)
+    }), Text({ content: "   peak ", fg: rgb(ctx.theme.dim) }), Text({ content: `${compactNumber(tokens.peakPerHour)}/h`, fg: rgb(ctx.theme.fg) }), Text({ content: "   avg ", fg: rgb(ctx.theme.dim) }), Text({ content: `${compactNumber(averagePerHour)}/h`, fg: rgb(ctx.theme.fg) }), Text({
       content: tokens.topModel === null ? "" : `   top ${tokens.topModel}`,
       fg: rgb(ctx.theme.faint)
     })));
@@ -15859,7 +15870,7 @@ function analyticsBody(ctx, analytics, timeframe) {
   const cols = process.stdout.columns ?? 80;
   const rows = process.stdout.rows ?? 24;
   const width = Math.max(24, Math.min(160, cols - 12));
-  const height = Math.max(4, Math.min(10, rows - 16));
+  const height = Math.max(4, Math.min(10, rows - 17));
   const tokens = analytics.tokens?.timeframes.find((entry) => entry.key === timeframe.key);
   return [timeframeBar(ctx, timeframe), throughputCard(ctx, tokens, timeframe, height, width)];
 }
@@ -17229,6 +17240,7 @@ function createUsageObserver(provider, contentType, onUsage) {
   let model = null;
   let input = 0;
   let output = 0;
+  let cacheRead = 0;
   let saw = false;
   const maxJson = 4000000;
   const consume = (text) => {
@@ -17242,7 +17254,8 @@ function createUsageObserver(provider, contentType, onUsage) {
       if (event.type === "message_start" && event.message) {
         model = event.message.model ?? model;
         const usage2 = event.message.usage ?? {};
-        input += (usage2.input_tokens ?? 0) + (usage2.cache_read_input_tokens ?? 0) + (usage2.cache_creation_input_tokens ?? 0);
+        input += (usage2.input_tokens ?? 0) + (usage2.cache_creation_input_tokens ?? 0);
+        cacheRead += usage2.cache_read_input_tokens ?? 0;
         saw = true;
       } else if (event.type === "message_delta" && event.usage) {
         output = event.usage.output_tokens ?? output;
@@ -17252,7 +17265,9 @@ function createUsageObserver(provider, contentType, onUsage) {
     }
     const usage = event.response?.usage ?? event.usage;
     if (usage) {
-      input = usage.input_tokens ?? usage.prompt_tokens ?? input;
+      const cached2 = usage.input_tokens_details?.cached_tokens ?? 0;
+      input = Math.max(0, (usage.input_tokens ?? usage.prompt_tokens ?? input) - cached2);
+      cacheRead = cached2;
       output = usage.output_tokens ?? usage.completion_tokens ?? output;
       model = event.response?.model ?? event.model ?? model;
       saw = true;
@@ -17263,8 +17278,9 @@ function createUsageObserver(provider, contentType, onUsage) {
       if (!isSse && jsonBuffer.length > 0) {
         consume(jsonBuffer);
       }
-      if (saw && input + output > 0) {
+      if (saw && input + output + cacheRead > 0) {
         onUsage({
+          cacheReadTokens: cacheRead,
           inputTokens: input,
           model: model && model.length > 0 ? model : null,
           outputTokens: output
@@ -17410,6 +17426,7 @@ function createProxyHandler(options) {
       if (options.record !== undefined && response.ok && forwarded.body !== null) {
         const observer = createUsageObserver(route.provider, response.headers.get("content-type") ?? "", (usage) => options.record?.({
           at: Date.now(),
+          cacheReadTokens: usage.cacheReadTokens,
           inputTokens: usage.inputTokens,
           model: usage.model,
           outputTokens: usage.outputTokens,
@@ -17635,6 +17652,7 @@ function selectRotation(input) {
 function priceTokenTimeframe(aggregate) {
   let totalInput = 0;
   let totalOutput = 0;
+  let totalCached = 0;
   let costTotal = 0;
   const byProvider = {
     anthropic: { costUsd: 0, tokens: 0 },
@@ -17642,10 +17660,11 @@ function priceTokenTimeframe(aggregate) {
   };
   let top = null;
   for (const entry of aggregate.byModel) {
-    const entryCost = costUsd(entry.model, entry.input, entry.output);
-    const entryTokens = entry.input + entry.output;
+    const entryCost = costUsd(entry.model, entry.input, entry.output, entry.cached);
+    const entryTokens = entry.input + entry.output + entry.cached;
     totalInput += entry.input;
     totalOutput += entry.output;
+    totalCached += entry.cached;
     costTotal += entryCost;
     const bucket = entry.provider === "anthropic" ? byProvider.anthropic : byProvider.openai;
     bucket.tokens += entryTokens;
@@ -17663,9 +17682,10 @@ function priceTokenTimeframe(aggregate) {
     key: aggregate.key,
     peakPerHour: peakBucket * (3600000 / aggregate.bucketMs),
     topModel: top?.model ?? null,
+    totalCached,
     totalInput,
     totalOutput,
-    totalTokens: totalInput + totalOutput
+    totalTokens: totalInput + totalOutput + totalCached
   };
 }
 var defaultDependencies = {
@@ -18246,6 +18266,7 @@ function migrate(database) {
       account_id TEXT,
       model TEXT,
       input_tokens INTEGER NOT NULL DEFAULT 0,
+      cache_read_tokens INTEGER NOT NULL DEFAULT 0,
       output_tokens INTEGER NOT NULL DEFAULT 0
     );
     CREATE INDEX IF NOT EXISTS token_events_at ON token_events(at);
@@ -18253,6 +18274,11 @@ function migrate(database) {
     DROP TABLE IF EXISTS runtime_sessions;
   `);
   const accountColumns = new Set(database.query("PRAGMA table_info(accounts)").all().map((column) => column.name));
+  const tokenColumns = new Set(database.query("PRAGMA table_info(token_events)").all().map((column) => column.name));
+  if (!tokenColumns.has("cache_read_tokens")) {
+    database.exec("ALTER TABLE token_events ADD COLUMN cache_read_tokens INTEGER NOT NULL DEFAULT 0");
+    database.exec("DELETE FROM token_events");
+  }
   const requiresLabelMigration = !accountColumns.has("label");
   const requiresExternalAccountIdMigration = !accountColumns.has("external_account_id");
   const requiresExternalUserIdMigration = !accountColumns.has("external_user_id");
@@ -18440,12 +18466,12 @@ function createStateStore(databasePath) {
   }
   function recordTokenEvent(event) {
     const parsed = TokenEventSchema.parse(event);
-    database.query("INSERT INTO token_events (at, provider, account_id, model, input_tokens, output_tokens) VALUES (?, ?, ?, ?, ?, ?)").run(parsed.at, parsed.provider, parsed.accountId, parsed.model, parsed.inputTokens, parsed.outputTokens);
+    database.query("INSERT INTO token_events (at, provider, account_id, model, input_tokens, output_tokens, cache_read_tokens) VALUES (?, ?, ?, ?, ?, ?, ?)").run(parsed.at, parsed.provider, parsed.accountId, parsed.model, parsed.inputTokens, parsed.outputTokens, parsed.cacheReadTokens);
     database.query("DELETE FROM token_events WHERE at < ?").run(parsed.at - maxTokenEventAgeMs);
   }
   function tokenAnalytics(nowMillis, timeframes) {
-    const bucketQuery = database.query("SELECT CAST((at - ?) / ? AS INTEGER) AS b, SUM(input_tokens + output_tokens) AS tokens FROM token_events WHERE at >= ? AND at <= ? GROUP BY b");
-    const modelQuery = database.query("SELECT model, provider, SUM(input_tokens) AS input, SUM(output_tokens) AS output FROM token_events WHERE at >= ? AND at <= ? GROUP BY model, provider");
+    const bucketQuery = database.query("SELECT CAST((at - ?) / ? AS INTEGER) AS b, SUM(input_tokens + output_tokens + cache_read_tokens) AS tokens FROM token_events WHERE at >= ? AND at <= ? GROUP BY b");
+    const modelQuery = database.query("SELECT model, provider, SUM(input_tokens) AS input, SUM(output_tokens) AS output, SUM(cache_read_tokens) AS cached FROM token_events WHERE at >= ? AND at <= ? GROUP BY model, provider");
     return timeframes.map(({ key, ms }) => {
       const start = nowMillis - ms;
       const bucketMs = Math.max(1, Math.round(ms / tokenBucketCount));
@@ -18455,6 +18481,7 @@ function createStateStore(databasePath) {
         buckets[index] = (buckets[index] ?? 0) + row.tokens;
       }
       const byModel = modelQuery.all(start, nowMillis).map((row) => ({
+        cached: row.cached,
         input: row.input,
         model: row.model ?? "unknown",
         output: row.output,

package/package.json

@@ -62,5 +62,5 @@
 		"post-commit": "bun x @rubriclab/package post-commit"
 	},
 	"type": "module",
-	"version": "0.0.3"
+	"version": "0.0.5"
 }