tokentracker-cli 0.40.0 → 0.41.0

package/src/lib/rollout.js

@@ -4420,26 +4420,42 @@ async function parseKimiCodeIncremental({ wireFiles, cursors, queuePath, onProgr
 
       recordsProcessed++;
 
-      // Anthropic-style usage: input_tokens already excludes cache. OpenAI-compat
-      // models instead fold cached reads into input_tokens and expose them via
-      // input_tokens_details.cached_tokens — mirror kimi-code's own extractUsage
-      // so we never double-count cache as fresh input.
-      const cacheCreation = toNonNegativeInt(usage.cache_creation_input_tokens);
+      // kimi-code's wire usage comes in two shapes across versions:
+      //  - camelCase (proto 0.6.0+, current): { inputOther, inputCacheRead,
+      //    inputCacheCreation, output } where inputOther is already fresh
+      //    (non-cached) input — this is `response.usage` straight from the LLM
+      //    adapter (verified in @moonshot-ai/kimi-code 0.6.0/0.7.0/0.9.0).
+      //  - Anthropic-style (older): { input_tokens, output_tokens,
+      //    cache_read_input_tokens, cache_creation_input_tokens }. OpenAI-compat
+      //    models fold cached reads into input_tokens and expose them via
+      //    input_tokens_details.cached_tokens — subtract so we never double-count.
+      // step.end and usage.record carry the SAME per-step usage object, so we
+      // read only step.end here (reading both would double-count ~2x).
+      let cacheCreation;
       let cacheRead;
       let input;
-      if (usage.cache_read_input_tokens != null) {
-        cacheRead = toNonNegativeInt(usage.cache_read_input_tokens);
-        input = toNonNegativeInt(usage.input_tokens);
+      let output;
+      if (usage.inputOther != null) {
+        input = toNonNegativeInt(usage.inputOther);
+        cacheRead = toNonNegativeInt(usage.inputCacheRead);
+        cacheCreation = toNonNegativeInt(usage.inputCacheCreation);
+        output = toNonNegativeInt(usage.output);
       } else {
-        const details =
-          usage.input_tokens_details && typeof usage.input_tokens_details === "object"
-            ? usage.input_tokens_details
-            : null;
-        const cached = toNonNegativeInt(details ? details.cached_tokens : 0);
-        cacheRead = cached;
-        input = Math.max(0, toNonNegativeInt(usage.input_tokens) - cached);
+        cacheCreation = toNonNegativeInt(usage.cache_creation_input_tokens);
+        if (usage.cache_read_input_tokens != null) {
+          cacheRead = toNonNegativeInt(usage.cache_read_input_tokens);
+          input = toNonNegativeInt(usage.input_tokens);
+        } else {
+          const details =
+            usage.input_tokens_details && typeof usage.input_tokens_details === "object"
+              ? usage.input_tokens_details
+              : null;
+          const cached = toNonNegativeInt(details ? details.cached_tokens : 0);
+          cacheRead = cached;
+          input = Math.max(0, toNonNegativeInt(usage.input_tokens) - cached);
+        }
+        output = toNonNegativeInt(usage.output_tokens);
       }
-      const output = toNonNegativeInt(usage.output_tokens);
       if (input === 0 && output === 0 && cacheRead === 0 && cacheCreation === 0) {
         seenIds.add(id);
         continue;

package/src/lib/usage-limits.js

@@ -21,6 +21,7 @@ const {
   extractCursorSessionToken,
   fetchCursorUsageSummary,
 } = require("./cursor-config");
+const { fetchGrokLimits } = require("./grok-limits");
 
 // 2-minute in-memory cache
 let cache = { data: null, fetchedAt: 0 };
@@ -1698,7 +1699,7 @@ async function getUsageLimits({
   const codexPlanType = codexAuthRefreshed?.planType || null;
 
   const providerFetch = withFetchTimeout(fetchImpl, providerTimeoutMs);
-  const [claudeResult, codexResult, cursor, kimi, gemini, kiro, antigravity, copilot] = await Promise.all([
+  const [claudeResult, codexResult, cursor, kimi, gemini, kiro, antigravity, copilot, grok] = await Promise.all([
     claudeToken
       ? withProviderTimeout(fetchClaudeUsageLimits(claudeToken, { fetchImpl: providerFetch, maxAttempts: 1 }), "Claude", providerTimeoutMs).then(
           (value) => ({ status: "fulfilled", value }),
@@ -1725,6 +1726,8 @@ async function getUsageLimits({
     fetchAntigravityLimits({ home, commandRunner, requestFn, nowMs }),
     withProviderTimeout(fetchCopilotLimits({ home, env, fetchImpl: providerFetch }), "GitHub Copilot", providerTimeoutMs)
       .catch((reason) => ({ configured: true, error: reason?.message || "Unknown error" })),
+    withProviderTimeout(fetchGrokLimits({ home, env, fetchImpl: providerFetch }), "Grok Build", providerTimeoutMs)
+      .catch((reason) => ({ configured: true, error: reason?.message || "Unknown error" })),
   ]);
 
   let claude;
@@ -1776,6 +1779,7 @@ async function getUsageLimits({
     kiro: withPlanLabel(kiro, kiro.plan_name, "Kiro"),
     antigravity: withPlanLabel(antigravity, antigravity.account_plan, "Antigravity"),
     copilot: withPlanLabel(copilot, copilot.plan_name, "Copilot"),
+    grok: withPlanLabel(grok, null, "Grok"),
   };
 
   cache = { data, fetchedAt: nowMs };
@@ -1803,4 +1807,5 @@ module.exports = {
   fetchCopilotLimits,
   readCopilotOauthToken,
   describeCopilotOtelStatus,
+  fetchGrokLimits,
 };