@chenpu17/cc-gw 0.4.2 → 0.5.0

package/src/server/dist/index.js

@@ -11103,7 +11103,11 @@ async function migrateDailyMetricsTable(db) {
   if (!hasEndpointColumn || !hasCompositePrimaryKey) {
     const endpointSelector = hasEndpointColumn ? "COALESCE(endpoint, 'anthropic')" : "'anthropic'";
     const hasCachedTokensColumn = columns.some((column) => column.name === "total_cached_tokens");
+    const hasCacheReadColumn = columns.some((column) => column.name === "total_cache_read_tokens");
+    const hasCacheCreationColumn = columns.some((column) => column.name === "total_cache_creation_tokens");
     const cachedTokensSelector = hasCachedTokensColumn ? "COALESCE(total_cached_tokens, 0)" : "0";
+    const cacheReadSelector = hasCacheReadColumn ? "COALESCE(total_cache_read_tokens, 0)" : "0";
+    const cacheCreationSelector = hasCacheCreationColumn ? "COALESCE(total_cache_creation_tokens, 0)" : "0";
     await exec(
       db,
       `ALTER TABLE daily_metrics RENAME TO daily_metrics_old;
@@ -11114,16 +11118,20 @@ async function migrateDailyMetricsTable(db) {
          total_input_tokens INTEGER DEFAULT 0,
          total_output_tokens INTEGER DEFAULT 0,
          total_cached_tokens INTEGER DEFAULT 0,
+         total_cache_read_tokens INTEGER DEFAULT 0,
+         total_cache_creation_tokens INTEGER DEFAULT 0,
          total_latency_ms INTEGER DEFAULT 0,
          PRIMARY KEY (date, endpoint)
        );
-       INSERT INTO daily_metrics (date, endpoint, request_count, total_input_tokens, total_output_tokens, total_cached_tokens, total_latency_ms)
+       INSERT INTO daily_metrics (date, endpoint, request_count, total_input_tokens, total_output_tokens, total_cached_tokens, total_cache_read_tokens, total_cache_creation_tokens, total_latency_ms)
          SELECT date,
                 ${endpointSelector},
                 request_count,
                 total_input_tokens,
                 total_output_tokens,
                 ${cachedTokensSelector},
+                ${cacheReadSelector},
+                ${cacheCreationSelector},
                 total_latency_ms
            FROM daily_metrics_old;
        DROP TABLE daily_metrics_old;`
@@ -11181,6 +11189,8 @@ async function ensureSchema(db) {
       total_input_tokens INTEGER DEFAULT 0,
       total_output_tokens INTEGER DEFAULT 0,
       total_cached_tokens INTEGER DEFAULT 0,
+      total_cache_read_tokens INTEGER DEFAULT 0,
+      total_cache_creation_tokens INTEGER DEFAULT 0,
       total_latency_ms INTEGER DEFAULT 0,
       PRIMARY KEY (date, endpoint)
     );
@@ -11217,6 +11227,8 @@ async function ensureSchema(db) {
   );
   await maybeAddColumn(db, "request_logs", "client_model", "TEXT");
   await maybeAddColumn(db, "request_logs", "cached_tokens", "INTEGER");
+  await maybeAddColumn(db, "request_logs", "cache_read_tokens", "INTEGER DEFAULT 0");
+  await maybeAddColumn(db, "request_logs", "cache_creation_tokens", "INTEGER DEFAULT 0");
   await maybeAddColumn(db, "request_logs", "ttft_ms", "INTEGER");
   await maybeAddColumn(db, "request_logs", "tpot_ms", "REAL");
   await maybeAddColumn(db, "request_logs", "stream", "INTEGER");
@@ -11239,6 +11251,8 @@ async function ensureSchema(db) {
   await maybeAddColumn(db, "api_keys", "total_output_tokens", "INTEGER DEFAULT 0");
   await migrateDailyMetricsTable(db);
   await maybeAddColumn(db, "daily_metrics", "total_cached_tokens", "INTEGER DEFAULT 0");
+  await maybeAddColumn(db, "daily_metrics", "total_cache_read_tokens", "INTEGER DEFAULT 0");
+  await maybeAddColumn(db, "daily_metrics", "total_cache_creation_tokens", "INTEGER DEFAULT 0");
   await run(db, "CREATE UNIQUE INDEX IF NOT EXISTS idx_api_keys_hash ON api_keys(key_hash) WHERE key_hash IS NOT NULL");
   await run(db, "UPDATE api_keys SET key_hash = '*' WHERE is_wildcard = 1 AND (key_hash IS NULL OR key_hash = '')");
   await run(db, "UPDATE api_keys SET updated_at = created_at WHERE updated_at IS NULL");
@@ -11404,6 +11418,14 @@ async function updateLogTokens(requestId, values) {
     values.outputTokens,
     values.cachedTokens ?? null
   ];
+  if (values.cacheReadTokens !== void 0) {
+    setters.push("cache_read_tokens = ?");
+    params.push(values.cacheReadTokens ?? null);
+  }
+  if (values.cacheCreationTokens !== void 0) {
+    setters.push("cache_creation_tokens = ?");
+    params.push(values.cacheCreationTokens ?? null);
+  }
   if (values.ttftMs !== void 0) {
     setters.push("ttft_ms = ?");
     params.push(values.ttftMs ?? null);
@@ -11455,25 +11477,33 @@ async function upsertLogPayload(requestId, payload) {
   );
 }
 async function updateMetrics(date, endpoint, delta) {
-  await runQuery(
-    `INSERT INTO daily_metrics (date, endpoint, request_count, total_input_tokens, total_output_tokens, total_cached_tokens, total_latency_ms)
-     VALUES (?, ?, ?, ?, ?, ?, ?)
-     ON CONFLICT(date, endpoint) DO UPDATE SET
-       request_count = daily_metrics.request_count + excluded.request_count,
-       total_input_tokens = daily_metrics.total_input_tokens + excluded.total_input_tokens,
-       total_output_tokens = daily_metrics.total_output_tokens + excluded.total_output_tokens,
-       total_cached_tokens = daily_metrics.total_cached_tokens + excluded.total_cached_tokens,
-       total_latency_ms = daily_metrics.total_latency_ms + excluded.total_latency_ms`,
-    [
-      date,
-      endpoint,
-      delta.requests,
-      delta.inputTokens,
-      delta.outputTokens,
-      delta.cachedTokens ?? 0,
-      delta.latencyMs
-    ]
-  );
+  try {
+    await runQuery(
+      `INSERT INTO daily_metrics (date, endpoint, request_count, total_input_tokens, total_output_tokens, total_cached_tokens, total_cache_read_tokens, total_cache_creation_tokens, total_latency_ms)
+       VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)
+       ON CONFLICT(date, endpoint) DO UPDATE SET
+         request_count = daily_metrics.request_count + excluded.request_count,
+         total_input_tokens = daily_metrics.total_input_tokens + excluded.total_input_tokens,
+         total_output_tokens = daily_metrics.total_output_tokens + excluded.total_output_tokens,
+         total_cached_tokens = daily_metrics.total_cached_tokens + excluded.total_cached_tokens,
+         total_cache_read_tokens = daily_metrics.total_cache_read_tokens + excluded.total_cache_read_tokens,
+         total_cache_creation_tokens = daily_metrics.total_cache_creation_tokens + excluded.total_cache_creation_tokens,
+         total_latency_ms = daily_metrics.total_latency_ms + excluded.total_latency_ms`,
+      [
+        date,
+        endpoint,
+        delta.requests,
+        delta.inputTokens,
+        delta.outputTokens,
+        delta.cachedTokens ?? 0,
+        delta.cacheReadTokens ?? 0,
+        delta.cacheCreationTokens ?? 0,
+        delta.latencyMs
+      ]
+    );
+  } catch (err) {
+    console.error("[updateMetrics] Failed to update metrics:", err);
+  }
 }
 
 // metrics/activity.ts
@@ -11943,23 +11973,24 @@ function computeTpot(totalLatencyMs, outputTokens, options) {
   return Number.isFinite(raw) ? roundTwoDecimals(raw) : null;
 }
 function resolveCachedTokens(usage) {
+  const result = { read: 0, creation: 0 };
   if (!usage || typeof usage !== "object") {
-    return null;
+    return result;
+  }
+  if (typeof usage.cache_read_input_tokens === "number") {
+    result.read = usage.cache_read_input_tokens;
+  }
+  if (typeof usage.cache_creation_input_tokens === "number") {
+    result.creation = usage.cache_creation_input_tokens;
   }
   if (typeof usage.cached_tokens === "number") {
-    return usage.cached_tokens;
+    result.read = usage.cached_tokens;
   }
   const promptDetails = usage.prompt_tokens_details;
   if (promptDetails && typeof promptDetails.cached_tokens === "number") {
-    return promptDetails.cached_tokens;
-  }
-  if (typeof usage.cache_read_input_tokens === "number") {
-    return usage.cache_read_input_tokens;
-  }
-  if (typeof usage.cache_creation_input_tokens === "number") {
-    return usage.cache_creation_input_tokens;
+    result.read = promptDetails.cached_tokens;
   }
-  return null;
+  return result;
 }
 function cloneOriginalPayload(value) {
   const structuredCloneFn = globalThis.structuredClone;
@@ -12222,7 +12253,8 @@ async function registerMessagesRoute(app) {
         if (providerType === "anthropic") {
           let inputTokens2 = json.usage?.input_tokens ?? 0;
           let outputTokens2 = json.usage?.output_tokens ?? 0;
-          const cachedTokens3 = resolveCachedTokens(json.usage);
+          const cached2 = resolveCachedTokens(json.usage);
+          const cachedTokens2 = cached2.read + cached2.creation;
           if (!inputTokens2) {
             inputTokens2 = target.tokenEstimate || estimateTokens(normalized, target.modelId);
           }
@@ -12233,13 +12265,13 @@ async function registerMessagesRoute(app) {
           logUsage("non_stream.anthropic", {
             input: inputTokens2,
             output: outputTokens2,
-            cached: cachedTokens3
+            cached: cachedTokens2
           });
           const latencyMs2 = Date.now() - requestStart;
           await updateLogTokens(logId, {
             inputTokens: inputTokens2,
             outputTokens: outputTokens2,
-            cachedTokens: cachedTokens3,
+            cachedTokens: cachedTokens2,
             ttftMs: latencyMs2,
             tpotMs: computeTpot(latencyMs2, outputTokens2, { streaming: false })
           });
@@ -12248,7 +12280,9 @@ async function registerMessagesRoute(app) {
             requests: 1,
             inputTokens: inputTokens2,
             outputTokens: outputTokens2,
-            cachedTokens: cachedTokens3,
+            cachedTokens: cachedTokens2,
+            cacheReadTokens: cached2.read,
+            cacheCreationTokens: cached2.creation,
             latencyMs: latencyMs2
           });
           if (storeResponsePayloads) {
@@ -12269,7 +12303,8 @@ async function registerMessagesRoute(app) {
         const claudeResponse = buildClaudeResponse(json, target.modelId);
         let inputTokens = json.usage?.prompt_tokens ?? 0;
         let outputTokens = json.usage?.completion_tokens ?? 0;
-        const cachedTokens2 = resolveCachedTokens(json.usage);
+        const cached = resolveCachedTokens(json.usage);
+        const cachedTokens = cached.read + cached.creation;
         if (!inputTokens) {
           inputTokens = target.tokenEstimate || estimateTokens(normalized, target.modelId);
         }
@@ -12280,13 +12315,13 @@ async function registerMessagesRoute(app) {
         logUsage("non_stream.openai", {
           input: inputTokens,
           output: outputTokens,
-          cached: cachedTokens2
+          cached: cachedTokens
         });
         const latencyMs = Date.now() - requestStart;
         await updateLogTokens(logId, {
           inputTokens,
           outputTokens,
-          cachedTokens: cachedTokens2,
+          cachedTokens,
           ttftMs: latencyMs,
           tpotMs: computeTpot(latencyMs, outputTokens, { streaming: false })
         });
@@ -12295,6 +12330,9 @@ async function registerMessagesRoute(app) {
           requests: 1,
           inputTokens,
           outputTokens,
+          cachedTokens,
+          cacheReadTokens: cached.read,
+          cacheCreationTokens: cached.creation,
           latencyMs
         });
         if (storeResponsePayloads) {
@@ -12466,9 +12504,7 @@ async function registerMessagesRoute(app) {
                         usagePrompt2 = payload2.usage.input_tokens ?? usagePrompt2;
                         usageCompletion2 = payload2.usage.output_tokens ?? usageCompletion2;
                         const maybeCached = resolveCachedTokens(payload2.usage);
-                        if (maybeCached !== null) {
-                          usageCached2 = maybeCached;
-                        }
+                        usageCached2 = maybeCached.read + maybeCached.creation;
                         lastUsagePayload = payload2.usage;
                       }
                       if (payload2?.delta) {
@@ -12495,9 +12531,7 @@ async function registerMessagesRoute(app) {
                         usagePrompt2 = payload2.usage.input_tokens ?? usagePrompt2;
                         usageCompletion2 = payload2.usage.output_tokens ?? usageCompletion2;
                         const maybeCached = resolveCachedTokens(payload2.usage);
-                        if (maybeCached !== null) {
-                          usageCached2 = maybeCached;
-                        }
+                        usageCached2 = maybeCached.read + maybeCached.creation;
                         lastUsagePayload = payload2.usage;
                       }
                       if (payload2?.stop_reason) {
@@ -12549,8 +12583,9 @@ async function registerMessagesRoute(app) {
         }
         const totalLatencyMs = Date.now() - requestStart;
         const ttftMs = firstTokenAt2 ? firstTokenAt2 - requestStart : null;
+        const cached = resolveCachedTokens(lastUsagePayload);
         if (usageCached2 === null) {
-          usageCached2 = resolveCachedTokens(lastUsagePayload);
+          usageCached2 = cached.read + cached.creation;
         }
         logUsage("stream.anthropic.final", {
           input: usagePrompt2,
@@ -12561,6 +12596,8 @@ async function registerMessagesRoute(app) {
           inputTokens: usagePrompt2,
           outputTokens: usageCompletion2,
           cachedTokens: usageCached2,
+          cacheReadTokens: cached.read,
+          cacheCreationTokens: cached.creation,
           ttftMs,
           tpotMs: computeTpot(totalLatencyMs, usageCompletion2, {
             streaming: true,
@@ -12573,6 +12610,8 @@ async function registerMessagesRoute(app) {
           inputTokens: usagePrompt2,
           outputTokens: usageCompletion2,
           cachedTokens: usageCached2,
+          cacheReadTokens: cached.read,
+          cacheCreationTokens: cached.creation,
           latencyMs: totalLatencyMs
         });
         if (storeResponsePayloads) {
@@ -12732,6 +12771,8 @@ data: ${JSON.stringify(data)}
               inputTokens: finalPromptTokens,
               outputTokens: finalCompletionTokens,
               cachedTokens: usageCached,
+              cacheReadTokens: 0,
+              cacheCreationTokens: 0,
               ttftMs,
               tpotMs: computeTpot(totalLatencyMs, finalCompletionTokens, {
                 streaming: true,
@@ -12743,7 +12784,9 @@ data: ${JSON.stringify(data)}
               requests: 1,
               inputTokens: finalPromptTokens,
               outputTokens: finalCompletionTokens,
-              cachedTokens: usageCached,
+              cachedTokens: usageCached ?? 0,
+              cacheReadTokens: 0,
+              cacheCreationTokens: 0,
               latencyMs: totalLatencyMs
             });
             if (storeResponsePayloads) {
@@ -12894,6 +12937,8 @@ data: ${JSON.stringify(data)}
           inputTokens: fallbackPrompt,
           outputTokens: fallbackCompletion,
           cachedTokens: usageCached,
+          cacheReadTokens: 0,
+          cacheCreationTokens: 0,
           ttftMs,
           tpotMs: computeTpot(totalLatencyMs, fallbackCompletion, {
             streaming: true,
@@ -12905,7 +12950,9 @@ data: ${JSON.stringify(data)}
           requests: 1,
           inputTokens: fallbackPrompt,
           outputTokens: fallbackCompletion,
-          cachedTokens: usageCached,
+          cachedTokens: usageCached ?? 0,
+          cacheReadTokens: 0,
+          cacheCreationTokens: 0,
           latencyMs: totalLatencyMs
         });
         if (storeResponsePayloads) {
@@ -13306,27 +13353,28 @@ function computeTpot2(totalLatencyMs, outputTokens, options) {
   return Number.isFinite(raw) ? roundTwoDecimals2(raw) : null;
 }
 function resolveCachedTokens2(usage) {
+  const result = { read: 0, creation: 0 };
   if (!usage || typeof usage !== "object") {
-    return null;
+    return result;
+  }
+  if (typeof usage.cache_read_input_tokens === "number") {
+    result.read = usage.cache_read_input_tokens;
+  }
+  if (typeof usage.cache_creation_input_tokens === "number") {
+    result.creation = usage.cache_creation_input_tokens;
   }
   if (typeof usage.cached_tokens === "number") {
-    return usage.cached_tokens;
+    result.read = usage.cached_tokens;
   }
   const promptDetails = usage.prompt_tokens_details;
   if (promptDetails && typeof promptDetails.cached_tokens === "number") {
-    return promptDetails.cached_tokens;
+    result.read = promptDetails.cached_tokens;
   }
   const inputDetails = usage.input_tokens_details;
   if (inputDetails && typeof inputDetails.cached_tokens === "number") {
-    return inputDetails.cached_tokens;
-  }
-  if (typeof usage.cache_read_input_tokens === "number") {
-    return usage.cache_read_input_tokens;
-  }
-  if (typeof usage.cache_creation_input_tokens === "number") {
-    return usage.cache_creation_input_tokens;
+    result.read = inputDetails.cached_tokens;
   }
-  return null;
+  return result;
 }
 var generateId = (prefix) => `${prefix}_${Math.random().toString(36).slice(2, 10)}`;
 var isText = (input) => typeof input === "string" && input.length > 0;
@@ -13848,17 +13896,20 @@ async function registerOpenAiRoutes(app) {
           if (!Number.isFinite(inputTokens3) || inputTokens3 <= 0) {
             inputTokens3 = target.tokenEstimate ?? estimateTokens(normalized, target.modelId);
           }
-          const cachedTokens3 = resolveCachedTokens2(usagePayload2);
+          const cached2 = resolveCachedTokens2(usagePayload2);
+          const cachedTokens2 = cached2.read + cached2.creation;
           const latencyMs3 = Date.now() - requestStart;
           const openAIResponse = buildOpenAIResponseFromClaude(parsed, target.modelId, converted, {
             inputTokens: inputTokens3,
             outputTokens: outputTokens3,
-            cachedTokens: cachedTokens3
+            cachedTokens: cachedTokens2
           });
           await updateLogTokens(logId, {
             inputTokens: inputTokens3,
             outputTokens: outputTokens3,
-            cachedTokens: cachedTokens3,
+            cachedTokens: usageCached,
+            cacheReadTokens: cached2.read,
+            cacheCreationTokens: cached2.creation,
             ttftMs: latencyMs3,
             tpotMs: computeTpot2(latencyMs3, outputTokens3, { streaming: false })
           });
@@ -13867,7 +13918,9 @@ async function registerOpenAiRoutes(app) {
             requests: 1,
             inputTokens: inputTokens3,
             outputTokens: outputTokens3,
-            cachedTokens: cachedTokens3,
+            cachedTokens: usageCached,
+            cacheReadTokens: usageCacheRead,
+            cacheCreationTokens: usageCacheCreation,
             latencyMs: latencyMs3
           });
           if (storeResponsePayloads) {
@@ -13902,12 +13955,15 @@ async function registerOpenAiRoutes(app) {
           return 0;
         })();
         const outputTokens2 = baseOutputTokens + reasoningTokens2;
-        const cachedTokens2 = resolveCachedTokens2(usagePayload);
+        const cached = resolveCachedTokens2(usagePayload);
+        const cachedTokens = cached.read + cached.creation;
         const latencyMs2 = Date.now() - requestStart;
         await updateLogTokens(logId, {
           inputTokens: inputTokens2,
           outputTokens: outputTokens2,
-          cachedTokens: cachedTokens2,
+          cachedTokens: usageCached,
+          cacheReadTokens: cached.read,
+          cacheCreationTokens: cached.creation,
           ttftMs: usagePayload?.first_token_latency_ms ?? latencyMs2,
           tpotMs: usagePayload?.tokens_per_second ? computeTpot2(latencyMs2, outputTokens2, { streaming: false, reasoningTokens: reasoningTokens2 }) : null
         });
@@ -13949,6 +14005,8 @@ async function registerOpenAiRoutes(app) {
         let usagePrompt2 = null;
         let usageCompletion2 = null;
         let usageCached2 = null;
+        let usageCacheRead2 = 0;
+        let usageCacheCreation2 = 0;
         let lastUsagePayload = null;
         let firstTokenAt2 = null;
         let claudeMessageId = null;
@@ -13990,9 +14048,9 @@ async function registerOpenAiRoutes(app) {
           );
           if (usageCached2 == null) {
             const candidate = resolveCachedTokens2(usagePayload);
-            if (candidate != null) {
-              usageCached2 = candidate;
-            }
+            usageCacheRead2 = candidate.read;
+            usageCacheCreation2 = candidate.creation;
+            usageCached2 = candidate.read + candidate.creation;
           }
           lastUsagePayload = usagePayload;
         };
@@ -14263,7 +14321,8 @@ async function registerOpenAiRoutes(app) {
         ensureCreatedSent();
         let finalPromptTokens = typeof usagePrompt2 === "number" && usagePrompt2 > 0 ? usagePrompt2 : target.tokenEstimate ?? estimateTokens(normalized, target.modelId);
         let finalCompletionTokens = typeof usageCompletion2 === "number" && usageCompletion2 > 0 ? usageCompletion2 : aggregatedText ? estimateTextTokens(aggregatedText, target.modelId) : 0;
-        const finalCachedTokens = usageCached2 != null ? usageCached2 : resolveCachedTokens2(lastUsagePayload);
+        const finalCachedResult = usageCached2 != null ? { read: usageCacheRead2, creation: usageCacheCreation2 } : resolveCachedTokens2(lastUsagePayload);
+        const finalCachedTokens = finalCachedResult.read + finalCachedResult.creation;
         const totalLatencyMs = Date.now() - requestStart;
         const ttftMs = firstTokenAt2 ? firstTokenAt2 - requestStart : null;
         const openAIResponse = buildOpenAIResponseFromClaude(claudeMessage, target.modelId, converted, {
@@ -14294,6 +14353,8 @@ async function registerOpenAiRoutes(app) {
           inputTokens: finalPromptTokens,
           outputTokens: finalCompletionTokens,
           cachedTokens: finalCachedTokens ?? null,
+          cacheReadTokens: 0,
+          cacheCreationTokens: 0,
           ttftMs,
           tpotMs: computeTpot2(totalLatencyMs, finalCompletionTokens, {
             streaming: true,
@@ -14305,7 +14366,9 @@ async function registerOpenAiRoutes(app) {
           requests: 1,
           inputTokens: finalPromptTokens,
           outputTokens: finalCompletionTokens,
-          cachedTokens: usageCached2,
+          cachedTokens: finalCachedTokens,
+          cacheReadTokens: finalCachedResult.read,
+          cacheCreationTokens: finalCachedResult.creation,
           latencyMs: totalLatencyMs
         });
         if (storeResponsePayloads && capturedResponseChunks2) {
@@ -14325,6 +14388,8 @@ async function registerOpenAiRoutes(app) {
       let usageCompletion = null;
       let usageReasoning = null;
       let usageCached = null;
+      let usageCacheRead = 0;
+      let usageCacheCreation = 0;
       let firstTokenAt = null;
       let chunkCount = 0;
       const capturedResponseChunks = storeResponsePayloads ? [] : null;
@@ -14364,7 +14429,10 @@ async function registerOpenAiRoutes(app) {
             usageReasoning
           );
           if (usageCached == null) {
-            usageCached = resolveCachedTokens2(usagePayload);
+            const cachedResult = resolveCachedTokens2(usagePayload);
+            usageCacheRead = cachedResult.read;
+            usageCacheCreation = cachedResult.creation;
+            usageCached = cachedResult.read + cachedResult.creation;
           }
           if (OPENAI_DEBUG) {
             debugLog("usage payload received", usagePayload);
@@ -14456,6 +14524,8 @@ async function registerOpenAiRoutes(app) {
         inputTokens,
         outputTokens,
         cachedTokens: usageCached,
+        cacheReadTokens: 0,
+        cacheCreationTokens: 0,
         ttftMs: firstTokenAt ? firstTokenAt - requestStart : null,
         tpotMs: computeTpot2(latencyMs, outputTokens, {
           streaming: true,
@@ -14713,12 +14783,15 @@ async function registerOpenAiRoutes(app) {
             inputTokens: inputTokens3,
             outputTokens: outputTokens3
           });
-          const cachedTokens3 = resolveCachedTokens2(usagePayload2);
+          const cached2 = resolveCachedTokens2(usagePayload2);
+          const cachedTokens2 = cached2.read + cached2.creation;
           const latencyMs3 = Date.now() - requestStart;
           await updateLogTokens(logId, {
             inputTokens: inputTokens3,
             outputTokens: outputTokens3,
-            cachedTokens: cachedTokens3,
+            cachedTokens: usageCached,
+            cacheReadTokens: cached2.read,
+            cacheCreationTokens: cached2.creation,
             ttftMs: latencyMs3,
             tpotMs: computeTpot2(latencyMs3, outputTokens3, { streaming: false })
           });
@@ -14727,7 +14800,9 @@ async function registerOpenAiRoutes(app) {
             requests: 1,
             inputTokens: inputTokens3,
             outputTokens: outputTokens3,
-            cachedTokens: cachedTokens3,
+            cachedTokens: usageCached,
+            cacheReadTokens: usageCacheRead,
+            cacheCreationTokens: usageCacheCreation,
             latencyMs: latencyMs3
           });
           if (storeResponsePayloads) {
@@ -14757,12 +14832,15 @@ async function registerOpenAiRoutes(app) {
           })(),
           target.modelId
         );
-        const cachedTokens2 = resolveCachedTokens2(usagePayload);
+        const cached = resolveCachedTokens2(usagePayload);
+        const cachedTokens = cached.read + cached.creation;
         const latencyMs2 = Date.now() - requestStart;
         await updateLogTokens(logId, {
           inputTokens: inputTokens2,
           outputTokens: outputTokens2,
-          cachedTokens: cachedTokens2,
+          cachedTokens: usageCached,
+          cacheReadTokens: cached.read,
+          cacheCreationTokens: cached.creation,
           ttftMs: usagePayload?.first_token_latency_ms ?? latencyMs2,
           tpotMs: usagePayload?.tokens_per_second ? computeTpot2(latencyMs2, outputTokens2, { streaming: false }) : null
         });
@@ -14804,6 +14882,8 @@ async function registerOpenAiRoutes(app) {
         let usagePrompt2 = null;
         let usageCompletion2 = null;
         let usageCached2 = null;
+        let usageCacheRead2 = 0;
+        let usageCacheCreation2 = 0;
         let lastUsagePayload = null;
         let firstTokenAt2 = null;
         let claudeStopReason = null;
@@ -14846,9 +14926,9 @@ async function registerOpenAiRoutes(app) {
           );
           if (usageCached2 == null) {
             const candidate = resolveCachedTokens2(usagePayload);
-            if (candidate != null) {
-              usageCached2 = candidate;
-            }
+            usageCacheRead2 = candidate.read;
+            usageCacheCreation2 = candidate.creation;
+            usageCached2 = candidate.read + candidate.creation;
           }
           lastUsagePayload = usagePayload;
         };
@@ -15159,7 +15239,8 @@ async function registerOpenAiRoutes(app) {
         }
         const finalPromptTokens = typeof usagePrompt2 === "number" && usagePrompt2 > 0 ? usagePrompt2 : target.tokenEstimate ?? estimateTokens(normalized, target.modelId);
         const finalCompletionTokens = typeof usageCompletion2 === "number" && usageCompletion2 > 0 ? usageCompletion2 : aggregatedText ? estimateTextTokens(aggregatedText, target.modelId) : 0;
-        const finalCachedTokens = usageCached2 != null ? usageCached2 : resolveCachedTokens2(lastUsagePayload);
+        const finalCachedResult = usageCached2 != null ? { read: usageCacheRead2, creation: usageCacheCreation2 } : resolveCachedTokens2(lastUsagePayload);
+        const finalCachedTokens = finalCachedResult.read + finalCachedResult.creation;
         const totalLatencyMs = Date.now() - requestStart;
         const ttftMs = firstTokenAt2 ? firstTokenAt2 - requestStart : null;
         const finishReason = mapClaudeStopReasonToChatFinish(claudeStopReason) ?? "stop";
@@ -15192,6 +15273,8 @@ async function registerOpenAiRoutes(app) {
           inputTokens: finalPromptTokens,
           outputTokens: finalCompletionTokens,
           cachedTokens: finalCachedTokens ?? null,
+          cacheReadTokens: 0,
+          cacheCreationTokens: 0,
           ttftMs,
           tpotMs: computeTpot2(totalLatencyMs, finalCompletionTokens, {
             streaming: true,
@@ -15203,7 +15286,9 @@ async function registerOpenAiRoutes(app) {
           requests: 1,
           inputTokens: finalPromptTokens,
           outputTokens: finalCompletionTokens,
-          cachedTokens: usageCached2,
+          cachedTokens: finalCachedTokens,
+          cacheReadTokens: finalCachedResult.read,
+          cacheCreationTokens: finalCachedResult.creation,
           latencyMs: totalLatencyMs
         });
         if (storeResponsePayloads && capturedResponseChunks2) {
@@ -15222,6 +15307,8 @@ async function registerOpenAiRoutes(app) {
       let usagePrompt = null;
       let usageCompletion = null;
       let usageCached = null;
+      let usageCacheRead = 0;
+      let usageCacheCreation = 0;
       let firstTokenAt = null;
       const capturedResponseChunks = storeResponsePayloads ? [] : null;
       const replyClosed = () => {
@@ -15254,7 +15341,10 @@ async function registerOpenAiRoutes(app) {
             usageCompletion
           );
           if (usageCached == null) {
-            usageCached = resolveCachedTokens2(usagePayload);
+            const cachedResult = resolveCachedTokens2(usagePayload);
+            usageCacheRead = cachedResult.read;
+            usageCacheCreation = cachedResult.creation;
+            usageCached = cachedResult.read + cachedResult.creation;
           }
         };
         while (true) {
@@ -15329,6 +15419,8 @@ async function registerOpenAiRoutes(app) {
         inputTokens,
         outputTokens,
         cachedTokens: usageCached,
+        cacheReadTokens: 0,
+        cacheCreationTokens: 0,
         ttftMs: firstTokenAt ? firstTokenAt - requestStart : null,
         tpotMs: computeTpot2(latencyMs, outputTokens, {
           streaming: true,
@@ -15614,6 +15706,8 @@ async function getDailyMetrics(days = 7, endpoint) {
             total_input_tokens AS inputTokens,
             total_output_tokens AS outputTokens,
             total_cached_tokens AS cachedTokens,
+            total_cache_read_tokens AS cacheReadTokens,
+            total_cache_creation_tokens AS cacheCreationTokens,
             total_latency_ms AS totalLatency
        FROM daily_metrics
        ${whereClause}
@@ -15627,6 +15721,8 @@ async function getDailyMetrics(days = 7, endpoint) {
     inputTokens: row.inputTokens ?? 0,
     outputTokens: row.outputTokens ?? 0,
     cachedTokens: row.cachedTokens ?? 0,
+    cacheReadTokens: row.cacheReadTokens ?? 0,
+    cacheCreationTokens: row.cacheCreationTokens ?? 0,
     avgLatencyMs: row.requestCount ? Math.round((row.totalLatency ?? 0) / row.requestCount) : 0
   })).reverse();
 }
@@ -15638,6 +15734,8 @@ async function getMetricsOverview(endpoint) {
        COALESCE(SUM(total_input_tokens), 0) AS inputTokens,
        COALESCE(SUM(total_output_tokens), 0) AS outputTokens,
        COALESCE(SUM(total_cached_tokens), 0) AS cachedTokens,
+       COALESCE(SUM(total_cache_read_tokens), 0) AS cacheReadTokens,
+       COALESCE(SUM(total_cache_creation_tokens), 0) AS cacheCreationTokens,
        COALESCE(SUM(total_latency_ms), 0) AS totalLatency
      FROM daily_metrics
      ${totalsWhere}`,
@@ -15649,6 +15747,8 @@ async function getMetricsOverview(endpoint) {
             total_input_tokens AS inputTokens,
             total_output_tokens AS outputTokens,
             total_cached_tokens AS cachedTokens,
+            total_cache_read_tokens AS cacheReadTokens,
+            total_cache_creation_tokens AS cacheCreationTokens,
             total_latency_ms AS totalLatency
        FROM daily_metrics
        WHERE date = ?
@@ -15666,6 +15766,8 @@ async function getMetricsOverview(endpoint) {
       inputTokens: totalsRow?.inputTokens ?? 0,
       outputTokens: totalsRow?.outputTokens ?? 0,
       cachedTokens: totalsRow?.cachedTokens ?? 0,
+      cacheReadTokens: totalsRow?.cacheReadTokens ?? 0,
+      cacheCreationTokens: totalsRow?.cacheCreationTokens ?? 0,
       avgLatencyMs: resolveAvg(totalsLatency, totalsRequests)
     },
     today: {
@@ -15673,6 +15775,8 @@ async function getMetricsOverview(endpoint) {
       inputTokens: todayRow?.inputTokens ?? 0,
       outputTokens: todayRow?.outputTokens ?? 0,
       cachedTokens: todayRow?.cachedTokens ?? 0,
+      cacheReadTokens: todayRow?.cacheReadTokens ?? 0,
+      cacheCreationTokens: todayRow?.cacheCreationTokens ?? 0,
       avgLatencyMs: resolveAvg(todayLatency, todayRequests)
     }
   };
@@ -17097,23 +17201,24 @@ function getPathsToRegister(basePath, protocol) {
   }
 }
 function resolveCachedTokens3(usage) {
+  const result = { read: 0, creation: 0 };
   if (!usage || typeof usage !== "object") {
-    return null;
+    return result;
+  }
+  if (typeof usage.cache_read_input_tokens === "number") {
+    result.read = usage.cache_read_input_tokens;
+  }
+  if (typeof usage.cache_creation_input_tokens === "number") {
+    result.creation = usage.cache_creation_input_tokens;
   }
   if (typeof usage.cached_tokens === "number") {
-    return usage.cached_tokens;
+    result.read = usage.cached_tokens;
   }
   const promptDetails = usage.prompt_tokens_details;
   if (promptDetails && typeof promptDetails.cached_tokens === "number") {
-    return promptDetails.cached_tokens;
-  }
-  if (typeof usage.cache_read_input_tokens === "number") {
-    return usage.cache_read_input_tokens;
-  }
-  if (typeof usage.cache_creation_input_tokens === "number") {
-    return usage.cache_creation_input_tokens;
+    result.read = promptDetails.cached_tokens;
   }
-  return null;
+  return result;
 }
 var roundTwoDecimals3 = (value) => Math.round(value * 100) / 100;
 function cloneOriginalPayload2(value) {
@@ -17470,12 +17575,15 @@ async function handleAnthropicProtocol(request, reply, endpoint, endpointId, app
       const json = await new Response(upstream.body).json();
       const inputTokens = json.usage?.input_tokens ?? estimateTokens(normalized, target.modelId);
       const outputTokens = json.usage?.output_tokens ?? 0;
-      const cachedTokens2 = resolveCachedTokens3(json.usage);
+      const cached = resolveCachedTokens3(json.usage);
+      const cachedTokens = cached.read + cached.creation;
       const latencyMs = Date.now() - requestStart;
       await updateLogTokens(logId, {
         inputTokens,
         outputTokens,
-        cachedTokens: cachedTokens2,
+        cachedTokens: usageCached,
+        cacheReadTokens: cached.read,
+        cacheCreationTokens: cached.creation,
         ttftMs: latencyMs,
         tpotMs: computeTpot3(latencyMs, outputTokens, { streaming: false })
       });
@@ -17484,7 +17592,9 @@ async function handleAnthropicProtocol(request, reply, endpoint, endpointId, app
         requests: 1,
         inputTokens,
         outputTokens,
-        cachedTokens: cachedTokens2,
+        cachedTokens: usageCached,
+        cacheReadTokens: usageCacheRead,
+        cacheCreationTokens: usageCacheCreation,
         latencyMs
       });
       if (storeResponsePayloads) {
@@ -17512,6 +17622,8 @@ async function handleAnthropicProtocol(request, reply, endpoint, endpointId, app
     let usagePrompt = 0;
     let usageCompletion = 0;
     let usageCached = null;
+    let usageCacheRead = 0;
+    let usageCacheCreation = 0;
     let firstTokenAt = null;
     const capturedChunks = storeResponsePayloads ? [] : null;
     try {
@@ -17541,9 +17653,9 @@ async function handleAnthropicProtocol(request, reply, endpoint, endpointId, app
                   usagePrompt = parsed.usage.input_tokens ?? usagePrompt;
                   usageCompletion = parsed.usage.output_tokens ?? usageCompletion;
                   const cached = resolveCachedTokens3(parsed.usage);
-                  if (cached !== null) {
-                    usageCached = cached;
-                  }
+                  usageCacheRead = cached.read;
+                  usageCacheCreation = cached.creation;
+                  usageCached = cached.read + cached.creation;
                 }
                 if (!firstTokenAt && (parsed?.type === "content_block_delta" || parsed?.delta?.text)) {
                   firstTokenAt = Date.now();
@@ -17570,6 +17682,8 @@ async function handleAnthropicProtocol(request, reply, endpoint, endpointId, app
       inputTokens: usagePrompt,
       outputTokens: usageCompletion,
       cachedTokens: usageCached,
+      cacheReadTokens: usageCacheRead,
+      cacheCreationTokens: usageCacheCreation,
       ttftMs,
       tpotMs: computeTpot3(totalLatencyMs, usageCompletion, {
         streaming: true,
@@ -17761,12 +17875,15 @@ async function handleOpenAIChatProtocol(request, reply, endpoint, endpointId, ap
       const usagePayload = json?.usage ?? null;
       const inputTokens2 = usagePayload?.prompt_tokens ?? usagePayload?.input_tokens ?? target.tokenEstimate ?? estimateTokens(normalized, target.modelId);
       const outputTokens2 = usagePayload?.completion_tokens ?? usagePayload?.output_tokens ?? estimateTextTokens(json?.choices?.[0]?.message?.content ?? "", target.modelId);
-      const cachedTokens2 = resolveCachedTokens3(usagePayload);
+      const cached = resolveCachedTokens3(usagePayload);
+      const cachedTokens = cached.read + cached.creation;
       const latencyMs2 = Date.now() - requestStart;
       await updateLogTokens(logId, {
         inputTokens: inputTokens2,
         outputTokens: outputTokens2,
-        cachedTokens: cachedTokens2,
+        cachedTokens: usageCached,
+        cacheReadTokens: cached.read,
+        cacheCreationTokens: cached.creation,
         ttftMs: latencyMs2,
         tpotMs: computeTpot3(latencyMs2, outputTokens2, { streaming: false })
       });
@@ -17775,7 +17892,9 @@ async function handleOpenAIChatProtocol(request, reply, endpoint, endpointId, ap
         requests: 1,
         inputTokens: inputTokens2,
         outputTokens: outputTokens2,
-        cachedTokens: cachedTokens2,
+        cachedTokens: usageCached,
+        cacheReadTokens: usageCacheRead,
+        cacheCreationTokens: usageCacheCreation,
         latencyMs: latencyMs2
       });
       if (storeResponsePayloads) {
@@ -17800,6 +17919,8 @@ async function handleOpenAIChatProtocol(request, reply, endpoint, endpointId, ap
     let usagePrompt = null;
     let usageCompletion = null;
     let usageCached = null;
+    let usageCacheRead = 0;
+    let usageCacheCreation = 0;
     let firstTokenAt = null;
     const capturedChunks = storeResponsePayloads ? [] : null;
     try {
@@ -17829,7 +17950,10 @@ async function handleOpenAIChatProtocol(request, reply, endpoint, endpointId, ap
                   if (usage) {
                     usagePrompt = usage.prompt_tokens ?? usage.input_tokens ?? usagePrompt;
                     usageCompletion = usage.completion_tokens ?? usage.output_tokens ?? usageCompletion;
-                    usageCached = usage.cached_tokens ?? usageCached;
+                    const cachedResult = resolveCachedTokens3(usage);
+                    usageCacheRead = cachedResult.read;
+                    usageCacheCreation = cachedResult.creation;
+                    usageCached = cachedResult.read + cachedResult.creation;
                   }
                 } catch {
                 }
@@ -17851,6 +17975,8 @@ async function handleOpenAIChatProtocol(request, reply, endpoint, endpointId, ap
       inputTokens,
       outputTokens,
       cachedTokens: usageCached,
+      cacheReadTokens: usageCacheRead,
+      cacheCreationTokens: usageCacheCreation,
       ttftMs: firstTokenAt ? firstTokenAt - requestStart : null,
       tpotMs: computeTpot3(latencyMs, outputTokens, {
         streaming: true,
@@ -18041,12 +18167,15 @@ async function handleOpenAIResponsesProtocol(request, reply, endpoint, endpointI
       const inputTokens2 = usagePayload?.prompt_tokens ?? usagePayload?.input_tokens ?? target.tokenEstimate ?? estimateTokens(normalized, target.modelId);
       const content = json?.response?.body?.content ?? json?.choices?.[0]?.message?.content ?? "";
       const outputTokens2 = usagePayload?.completion_tokens ?? usagePayload?.output_tokens ?? estimateTextTokens(content, target.modelId);
-      const cachedTokens2 = resolveCachedTokens3(usagePayload);
+      const cached = resolveCachedTokens3(usagePayload);
+      const cachedTokens = cached.read + cached.creation;
       const latencyMs2 = Date.now() - requestStart;
       await updateLogTokens(logId, {
         inputTokens: inputTokens2,
         outputTokens: outputTokens2,
-        cachedTokens: cachedTokens2,
+        cachedTokens: usageCached,
+        cacheReadTokens: cached.read,
+        cacheCreationTokens: cached.creation,
         ttftMs: latencyMs2,
         tpotMs: computeTpot3(latencyMs2, outputTokens2, { streaming: false })
       });
@@ -18055,7 +18184,9 @@ async function handleOpenAIResponsesProtocol(request, reply, endpoint, endpointI
         requests: 1,
         inputTokens: inputTokens2,
         outputTokens: outputTokens2,
-        cachedTokens: cachedTokens2,
+        cachedTokens: usageCached,
+        cacheReadTokens: usageCacheRead,
+        cacheCreationTokens: usageCacheCreation,
         latencyMs: latencyMs2
       });
       if (storeResponsePayloads) {
@@ -18080,6 +18211,8 @@ async function handleOpenAIResponsesProtocol(request, reply, endpoint, endpointI
     let usagePrompt = null;
     let usageCompletion = null;
     let usageCached = null;
+    let usageCacheRead = 0;
+    let usageCacheCreation = 0;
     let firstTokenAt = null;
     const capturedChunks = storeResponsePayloads ? [] : null;
     try {
@@ -18109,7 +18242,10 @@ async function handleOpenAIResponsesProtocol(request, reply, endpoint, endpointI
                   if (usage) {
                     usagePrompt = usage.prompt_tokens ?? usage.input_tokens ?? usagePrompt;
                     usageCompletion = usage.completion_tokens ?? usage.output_tokens ?? usageCompletion;
-                    usageCached = usage.cached_tokens ?? usageCached;
+                    const cachedResult = resolveCachedTokens3(usage);
+                    usageCacheRead = cachedResult.read;
+                    usageCacheCreation = cachedResult.creation;
+                    usageCached = cachedResult.read + cachedResult.creation;
                   }
                 } catch {
                 }
@@ -18131,6 +18267,8 @@ async function handleOpenAIResponsesProtocol(request, reply, endpoint, endpointI
       inputTokens,
       outputTokens,
       cachedTokens: usageCached,
+      cacheReadTokens: usageCacheRead,
+      cacheCreationTokens: usageCacheCreation,
       ttftMs: firstTokenAt ? firstTokenAt - requestStart : null,
       tpotMs: computeTpot3(latencyMs, outputTokens, {
         streaming: true,