@chenpu17/cc-gw 0.4.0 → 0.4.2

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@chenpu17/cc-gw",
-  "version": "0.4.0",
+  "version": "0.4.2",
   "private": false,
   "type": "module",
   "scripts": {

package/src/server/dist/index.js

@@ -11102,6 +11102,8 @@ async function migrateDailyMetricsTable(db) {
   const hasCompositePrimaryKey = primaryKeyColumns.length > 1;
   if (!hasEndpointColumn || !hasCompositePrimaryKey) {
     const endpointSelector = hasEndpointColumn ? "COALESCE(endpoint, 'anthropic')" : "'anthropic'";
+    const hasCachedTokensColumn = columns.some((column) => column.name === "total_cached_tokens");
+    const cachedTokensSelector = hasCachedTokensColumn ? "COALESCE(total_cached_tokens, 0)" : "0";
     await exec(
       db,
       `ALTER TABLE daily_metrics RENAME TO daily_metrics_old;
@@ -11111,15 +11113,17 @@ async function migrateDailyMetricsTable(db) {
          request_count INTEGER DEFAULT 0,
          total_input_tokens INTEGER DEFAULT 0,
          total_output_tokens INTEGER DEFAULT 0,
+         total_cached_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_latency_ms)
+       INSERT INTO daily_metrics (date, endpoint, request_count, total_input_tokens, total_output_tokens, total_cached_tokens, total_latency_ms)
          SELECT date,
                 ${endpointSelector},
                 request_count,
                 total_input_tokens,
                 total_output_tokens,
+                ${cachedTokensSelector},
                 total_latency_ms
            FROM daily_metrics_old;
        DROP TABLE daily_metrics_old;`
@@ -11176,6 +11180,7 @@ async function ensureSchema(db) {
       request_count INTEGER DEFAULT 0,
       total_input_tokens INTEGER DEFAULT 0,
       total_output_tokens INTEGER DEFAULT 0,
+      total_cached_tokens INTEGER DEFAULT 0,
       total_latency_ms INTEGER DEFAULT 0,
       PRIMARY KEY (date, endpoint)
     );
@@ -11233,6 +11238,7 @@ async function ensureSchema(db) {
   await maybeAddColumn(db, "api_keys", "total_input_tokens", "INTEGER DEFAULT 0");
   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 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");
@@ -11450,12 +11456,13 @@ 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_latency_ms)
-     VALUES (?, ?, ?, ?, ?, ?)
+    `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,
@@ -11463,6 +11470,7 @@ async function updateMetrics(date, endpoint, delta) {
       delta.requests,
       delta.inputTokens,
       delta.outputTokens,
+      delta.cachedTokens ?? 0,
       delta.latencyMs
     ]
   );
@@ -12214,7 +12222,7 @@ async function registerMessagesRoute(app) {
         if (providerType === "anthropic") {
           let inputTokens2 = json.usage?.input_tokens ?? 0;
           let outputTokens2 = json.usage?.output_tokens ?? 0;
-          const cachedTokens2 = resolveCachedTokens(json.usage);
+          const cachedTokens3 = resolveCachedTokens(json.usage);
           if (!inputTokens2) {
             inputTokens2 = target.tokenEstimate || estimateTokens(normalized, target.modelId);
           }
@@ -12225,13 +12233,13 @@ async function registerMessagesRoute(app) {
           logUsage("non_stream.anthropic", {
             input: inputTokens2,
             output: outputTokens2,
-            cached: cachedTokens2
+            cached: cachedTokens3
           });
           const latencyMs2 = Date.now() - requestStart;
           await updateLogTokens(logId, {
             inputTokens: inputTokens2,
             outputTokens: outputTokens2,
-            cachedTokens: cachedTokens2,
+            cachedTokens: cachedTokens3,
             ttftMs: latencyMs2,
             tpotMs: computeTpot(latencyMs2, outputTokens2, { streaming: false })
           });
@@ -12240,6 +12248,7 @@ async function registerMessagesRoute(app) {
             requests: 1,
             inputTokens: inputTokens2,
             outputTokens: outputTokens2,
+            cachedTokens: cachedTokens3,
             latencyMs: latencyMs2
           });
           if (storeResponsePayloads) {
@@ -12260,7 +12269,7 @@ 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 cachedTokens = resolveCachedTokens(json.usage);
+        const cachedTokens2 = resolveCachedTokens(json.usage);
         if (!inputTokens) {
           inputTokens = target.tokenEstimate || estimateTokens(normalized, target.modelId);
         }
@@ -12271,13 +12280,13 @@ async function registerMessagesRoute(app) {
         logUsage("non_stream.openai", {
           input: inputTokens,
           output: outputTokens,
-          cached: cachedTokens
+          cached: cachedTokens2
         });
         const latencyMs = Date.now() - requestStart;
         await updateLogTokens(logId, {
           inputTokens,
           outputTokens,
-          cachedTokens,
+          cachedTokens: cachedTokens2,
           ttftMs: latencyMs,
           tpotMs: computeTpot(latencyMs, outputTokens, { streaming: false })
         });
@@ -12563,6 +12572,7 @@ async function registerMessagesRoute(app) {
           requests: 1,
           inputTokens: usagePrompt2,
           outputTokens: usageCompletion2,
+          cachedTokens: usageCached2,
           latencyMs: totalLatencyMs
         });
         if (storeResponsePayloads) {
@@ -12733,6 +12743,7 @@ data: ${JSON.stringify(data)}
               requests: 1,
               inputTokens: finalPromptTokens,
               outputTokens: finalCompletionTokens,
+              cachedTokens: usageCached,
               latencyMs: totalLatencyMs
             });
             if (storeResponsePayloads) {
@@ -12894,6 +12905,7 @@ data: ${JSON.stringify(data)}
           requests: 1,
           inputTokens: fallbackPrompt,
           outputTokens: fallbackCompletion,
+          cachedTokens: usageCached,
           latencyMs: totalLatencyMs
         });
         if (storeResponsePayloads) {
@@ -13836,17 +13848,17 @@ async function registerOpenAiRoutes(app) {
           if (!Number.isFinite(inputTokens3) || inputTokens3 <= 0) {
             inputTokens3 = target.tokenEstimate ?? estimateTokens(normalized, target.modelId);
           }
-          const cachedTokens2 = resolveCachedTokens2(usagePayload2);
+          const cachedTokens3 = resolveCachedTokens2(usagePayload2);
           const latencyMs3 = Date.now() - requestStart;
           const openAIResponse = buildOpenAIResponseFromClaude(parsed, target.modelId, converted, {
             inputTokens: inputTokens3,
             outputTokens: outputTokens3,
-            cachedTokens: cachedTokens2
+            cachedTokens: cachedTokens3
           });
           await updateLogTokens(logId, {
             inputTokens: inputTokens3,
             outputTokens: outputTokens3,
-            cachedTokens: cachedTokens2,
+            cachedTokens: cachedTokens3,
             ttftMs: latencyMs3,
             tpotMs: computeTpot2(latencyMs3, outputTokens3, { streaming: false })
           });
@@ -13855,6 +13867,7 @@ async function registerOpenAiRoutes(app) {
             requests: 1,
             inputTokens: inputTokens3,
             outputTokens: outputTokens3,
+            cachedTokens: cachedTokens3,
             latencyMs: latencyMs3
           });
           if (storeResponsePayloads) {
@@ -13889,12 +13902,12 @@ async function registerOpenAiRoutes(app) {
           return 0;
         })();
         const outputTokens2 = baseOutputTokens + reasoningTokens2;
-        const cachedTokens = resolveCachedTokens2(usagePayload);
+        const cachedTokens2 = resolveCachedTokens2(usagePayload);
         const latencyMs2 = Date.now() - requestStart;
         await updateLogTokens(logId, {
           inputTokens: inputTokens2,
           outputTokens: outputTokens2,
-          cachedTokens,
+          cachedTokens: cachedTokens2,
           ttftMs: usagePayload?.first_token_latency_ms ?? latencyMs2,
           tpotMs: usagePayload?.tokens_per_second ? computeTpot2(latencyMs2, outputTokens2, { streaming: false, reasoningTokens: reasoningTokens2 }) : null
         });
@@ -14292,6 +14305,7 @@ async function registerOpenAiRoutes(app) {
           requests: 1,
           inputTokens: finalPromptTokens,
           outputTokens: finalCompletionTokens,
+          cachedTokens: usageCached2,
           latencyMs: totalLatencyMs
         });
         if (storeResponsePayloads && capturedResponseChunks2) {
@@ -14699,12 +14713,12 @@ async function registerOpenAiRoutes(app) {
             inputTokens: inputTokens3,
             outputTokens: outputTokens3
           });
-          const cachedTokens2 = resolveCachedTokens2(usagePayload2);
+          const cachedTokens3 = resolveCachedTokens2(usagePayload2);
           const latencyMs3 = Date.now() - requestStart;
           await updateLogTokens(logId, {
             inputTokens: inputTokens3,
             outputTokens: outputTokens3,
-            cachedTokens: cachedTokens2,
+            cachedTokens: cachedTokens3,
             ttftMs: latencyMs3,
             tpotMs: computeTpot2(latencyMs3, outputTokens3, { streaming: false })
           });
@@ -14713,6 +14727,7 @@ async function registerOpenAiRoutes(app) {
             requests: 1,
             inputTokens: inputTokens3,
             outputTokens: outputTokens3,
+            cachedTokens: cachedTokens3,
             latencyMs: latencyMs3
           });
           if (storeResponsePayloads) {
@@ -14742,12 +14757,12 @@ async function registerOpenAiRoutes(app) {
           })(),
           target.modelId
         );
-        const cachedTokens = resolveCachedTokens2(usagePayload);
+        const cachedTokens2 = resolveCachedTokens2(usagePayload);
         const latencyMs2 = Date.now() - requestStart;
         await updateLogTokens(logId, {
           inputTokens: inputTokens2,
           outputTokens: outputTokens2,
-          cachedTokens,
+          cachedTokens: cachedTokens2,
           ttftMs: usagePayload?.first_token_latency_ms ?? latencyMs2,
           tpotMs: usagePayload?.tokens_per_second ? computeTpot2(latencyMs2, outputTokens2, { streaming: false }) : null
         });
@@ -15188,6 +15203,7 @@ async function registerOpenAiRoutes(app) {
           requests: 1,
           inputTokens: finalPromptTokens,
           outputTokens: finalCompletionTokens,
+          cachedTokens: usageCached2,
           latencyMs: totalLatencyMs
         });
         if (storeResponsePayloads && capturedResponseChunks2) {
@@ -15597,6 +15613,7 @@ async function getDailyMetrics(days = 7, endpoint) {
             request_count AS requestCount,
             total_input_tokens AS inputTokens,
             total_output_tokens AS outputTokens,
+            total_cached_tokens AS cachedTokens,
             total_latency_ms AS totalLatency
        FROM daily_metrics
        ${whereClause}
@@ -15609,6 +15626,7 @@ async function getDailyMetrics(days = 7, endpoint) {
     requestCount: row.requestCount ?? 0,
     inputTokens: row.inputTokens ?? 0,
     outputTokens: row.outputTokens ?? 0,
+    cachedTokens: row.cachedTokens ?? 0,
     avgLatencyMs: row.requestCount ? Math.round((row.totalLatency ?? 0) / row.requestCount) : 0
   })).reverse();
 }
@@ -15619,6 +15637,7 @@ async function getMetricsOverview(endpoint) {
        COALESCE(SUM(request_count), 0) AS requests,
        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_latency_ms), 0) AS totalLatency
      FROM daily_metrics
      ${totalsWhere}`,
@@ -15629,6 +15648,7 @@ async function getMetricsOverview(endpoint) {
     `SELECT request_count AS requests,
             total_input_tokens AS inputTokens,
             total_output_tokens AS outputTokens,
+            total_cached_tokens AS cachedTokens,
             total_latency_ms AS totalLatency
        FROM daily_metrics
        WHERE date = ?
@@ -15645,12 +15665,14 @@ async function getMetricsOverview(endpoint) {
       requests: totalsRequests,
       inputTokens: totalsRow?.inputTokens ?? 0,
       outputTokens: totalsRow?.outputTokens ?? 0,
+      cachedTokens: totalsRow?.cachedTokens ?? 0,
       avgLatencyMs: resolveAvg(totalsLatency, totalsRequests)
     },
     today: {
       requests: todayRequests,
       inputTokens: todayRow?.inputTokens ?? 0,
       outputTokens: todayRow?.outputTokens ?? 0,
+      cachedTokens: todayRow?.cachedTokens ?? 0,
       avgLatencyMs: resolveAvg(todayLatency, todayRequests)
     }
   };
@@ -17094,6 +17116,13 @@ function resolveCachedTokens3(usage) {
   return null;
 }
 var roundTwoDecimals3 = (value) => Math.round(value * 100) / 100;
+function cloneOriginalPayload2(value) {
+  const structuredCloneFn = globalThis.structuredClone;
+  if (structuredCloneFn) {
+    return structuredCloneFn(value);
+  }
+  return JSON.parse(JSON.stringify(value));
+}
 function computeTpot3(totalLatencyMs, outputTokens, options) {
   if (!Number.isFinite(outputTokens) || outputTokens <= 0) {
     return null;
@@ -17230,7 +17259,8 @@ async function registerUnifiedHandler(app, path5, endpointId) {
     if (endpoint.paths && Array.isArray(endpoint.paths) && endpoint.paths.length > 0) {
       const matchedPath = endpoint.paths.find((p) => {
         const normalizedPath = p.path.startsWith("/") ? p.path : `/${p.path}`;
-        return actualPath === normalizedPath;
+        const expandedPaths = getPathsToRegister(normalizedPath, p.protocol);
+        return expandedPaths.includes(actualPath);
       });
       protocol = matchedPath?.protocol;
     } else if (endpoint.protocol) {
@@ -17268,6 +17298,37 @@ async function handleAnthropicProtocol(request, reply, endpoint, endpointId, app
     reply.code(400);
     return { error: "Invalid request body" };
   }
+  const rawUrl = typeof request.raw?.url === "string" ? request.raw.url : request.url ?? "";
+  let querySuffix = null;
+  if (typeof rawUrl === "string") {
+    const questionIndex = rawUrl.indexOf("?");
+    if (questionIndex !== -1) {
+      querySuffix = rawUrl.slice(questionIndex + 1);
+    }
+  }
+  if (!querySuffix && typeof request.querystring === "string") {
+    querySuffix = request.querystring || null;
+  }
+  const providerHeaders = {};
+  const headersToForward = [
+    "anthropic-version",
+    "anthropic-beta",
+    "x-stainless-arch",
+    "x-stainless-async",
+    "x-stainless-lang",
+    "x-stainless-os",
+    "x-stainless-package-version",
+    "x-stainless-runtime",
+    "x-stainless-runtime-version"
+  ];
+  for (const key of headersToForward) {
+    const value = request.headers[key];
+    if (typeof value === "string" && value.length > 0) {
+      providerHeaders[key] = value;
+    } else if (Array.isArray(value) && value.length > 0 && typeof value[0] === "string") {
+      providerHeaders[key] = value[0];
+    }
+  }
   const providedApiKey = extractApiKeyFromRequest2(request);
   let apiKeyContext;
   try {
@@ -17351,9 +17412,36 @@ async function handleAnthropicProtocol(request, reply, endpoint, endpointId, app
     const providerType = target.provider.type ?? "custom";
     const connector = getConnector(target.providerId);
     let providerBody;
+    let finalHeaders = providerHeaders;
     if (providerType === "anthropic") {
-      providerBody = { ...payload };
+      providerBody = cloneOriginalPayload2(payload);
       providerBody.model = target.modelId;
+      if (normalized.stream !== void 0) {
+        providerBody.stream = normalized.stream;
+      }
+      const collected = {};
+      const skip = /* @__PURE__ */ new Set(["content-length", "host", "connection", "transfer-encoding"]);
+      const sourceHeaders = request.raw?.headers ?? request.headers;
+      for (const [headerKey, headerValue] of Object.entries(sourceHeaders)) {
+        const lower = headerKey.toLowerCase();
+        if (skip.has(lower))
+          continue;
+        let value;
+        if (typeof headerValue === "string") {
+          value = headerValue;
+        } else if (Array.isArray(headerValue)) {
+          value = headerValue.find((item) => typeof item === "string" && item.length > 0);
+        }
+        if (value && value.length > 0) {
+          collected[lower] = value;
+        }
+      }
+      if (!("content-type" in collected)) {
+        collected["content-type"] = "application/json";
+      }
+      if (Object.keys(collected).length > 0) {
+        finalHeaders = collected;
+      }
     } else {
       providerBody = buildProviderBody(normalized, {
         maxTokens: payload.max_tokens,
@@ -17363,7 +17451,9 @@ async function handleAnthropicProtocol(request, reply, endpoint, endpointId, app
     const upstream = await connector.send({
       model: target.modelId,
       body: providerBody,
-      stream: normalized.stream
+      stream: normalized.stream,
+      query: querySuffix,
+      headers: finalHeaders
     });
     if (upstream.status >= 400) {
       reply.code(upstream.status);
@@ -17380,12 +17470,12 @@ 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 cachedTokens = resolveCachedTokens3(json.usage);
+      const cachedTokens2 = resolveCachedTokens3(json.usage);
       const latencyMs = Date.now() - requestStart;
       await updateLogTokens(logId, {
         inputTokens,
         outputTokens,
-        cachedTokens,
+        cachedTokens: cachedTokens2,
         ttftMs: latencyMs,
         tpotMs: computeTpot3(latencyMs, outputTokens, { streaming: false })
       });
@@ -17394,6 +17484,7 @@ async function handleAnthropicProtocol(request, reply, endpoint, endpointId, app
         requests: 1,
         inputTokens,
         outputTokens,
+        cachedTokens: cachedTokens2,
         latencyMs
       });
       if (storeResponsePayloads) {
@@ -17490,6 +17581,7 @@ async function handleAnthropicProtocol(request, reply, endpoint, endpointId, app
       requests: 1,
       inputTokens: usagePrompt,
       outputTokens: usageCompletion,
+      cachedTokens: usageCached,
       latencyMs: totalLatencyMs
     });
     if (storeResponsePayloads && capturedChunks) {
@@ -17522,6 +17614,37 @@ async function handleOpenAIChatProtocol(request, reply, endpoint, endpointId, ap
     reply.code(400);
     return { error: "Invalid request body" };
   }
+  const rawUrl = typeof request.raw?.url === "string" ? request.raw.url : request.url ?? "";
+  let querySuffix = null;
+  if (typeof rawUrl === "string") {
+    const questionIndex = rawUrl.indexOf("?");
+    if (questionIndex !== -1) {
+      querySuffix = rawUrl.slice(questionIndex + 1);
+    }
+  }
+  if (!querySuffix && typeof request.querystring === "string") {
+    querySuffix = request.querystring || null;
+  }
+  const providerHeaders = {};
+  const headersToForward = [
+    "anthropic-version",
+    "anthropic-beta",
+    "x-stainless-arch",
+    "x-stainless-async",
+    "x-stainless-lang",
+    "x-stainless-os",
+    "x-stainless-package-version",
+    "x-stainless-runtime",
+    "x-stainless-runtime-version"
+  ];
+  for (const key of headersToForward) {
+    const value = request.headers[key];
+    if (typeof value === "string" && value.length > 0) {
+      providerHeaders[key] = value;
+    } else if (Array.isArray(value) && value.length > 0 && typeof value[0] === "string") {
+      providerHeaders[key] = value[0];
+    }
+  }
   const providedApiKey = extractApiKeyFromRequest2(request);
   let apiKeyContext;
   try {
@@ -17618,7 +17741,9 @@ async function handleOpenAIChatProtocol(request, reply, endpoint, endpointId, ap
     const upstream = await connector.send({
       model: target.modelId,
       body: providerBody,
-      stream: normalized.stream
+      stream: normalized.stream,
+      query: querySuffix,
+      headers: providerHeaders
     });
     if (upstream.status >= 400) {
       reply.code(upstream.status);
@@ -17636,12 +17761,12 @@ 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 cachedTokens = resolveCachedTokens3(usagePayload);
+      const cachedTokens2 = resolveCachedTokens3(usagePayload);
       const latencyMs2 = Date.now() - requestStart;
       await updateLogTokens(logId, {
         inputTokens: inputTokens2,
         outputTokens: outputTokens2,
-        cachedTokens,
+        cachedTokens: cachedTokens2,
         ttftMs: latencyMs2,
         tpotMs: computeTpot3(latencyMs2, outputTokens2, { streaming: false })
       });
@@ -17650,6 +17775,7 @@ async function handleOpenAIChatProtocol(request, reply, endpoint, endpointId, ap
         requests: 1,
         inputTokens: inputTokens2,
         outputTokens: outputTokens2,
+        cachedTokens: cachedTokens2,
         latencyMs: latencyMs2
       });
       if (storeResponsePayloads) {
@@ -17767,6 +17893,37 @@ async function handleOpenAIResponsesProtocol(request, reply, endpoint, endpointI
     reply.code(400);
     return { error: "Invalid request body" };
   }
+  const rawUrl = typeof request.raw?.url === "string" ? request.raw.url : request.url ?? "";
+  let querySuffix = null;
+  if (typeof rawUrl === "string") {
+    const questionIndex = rawUrl.indexOf("?");
+    if (questionIndex !== -1) {
+      querySuffix = rawUrl.slice(questionIndex + 1);
+    }
+  }
+  if (!querySuffix && typeof request.querystring === "string") {
+    querySuffix = request.querystring || null;
+  }
+  const providerHeaders = {};
+  const headersToForward = [
+    "anthropic-version",
+    "anthropic-beta",
+    "x-stainless-arch",
+    "x-stainless-async",
+    "x-stainless-lang",
+    "x-stainless-os",
+    "x-stainless-package-version",
+    "x-stainless-runtime",
+    "x-stainless-runtime-version"
+  ];
+  for (const key of headersToForward) {
+    const value = request.headers[key];
+    if (typeof value === "string" && value.length > 0) {
+      providerHeaders[key] = value;
+    } else if (Array.isArray(value) && value.length > 0 && typeof value[0] === "string") {
+      providerHeaders[key] = value[0];
+    }
+  }
   const providedApiKey = extractApiKeyFromRequest2(request);
   let apiKeyContext;
   try {
@@ -17863,7 +18020,9 @@ async function handleOpenAIResponsesProtocol(request, reply, endpoint, endpointI
     const upstream = await connector.send({
       model: target.modelId,
       body: providerBody,
-      stream: normalized.stream
+      stream: normalized.stream,
+      query: querySuffix,
+      headers: providerHeaders
     });
     if (upstream.status >= 400) {
       reply.code(upstream.status);
@@ -17882,12 +18041,12 @@ 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 cachedTokens = resolveCachedTokens3(usagePayload);
+      const cachedTokens2 = resolveCachedTokens3(usagePayload);
       const latencyMs2 = Date.now() - requestStart;
       await updateLogTokens(logId, {
         inputTokens: inputTokens2,
         outputTokens: outputTokens2,
-        cachedTokens,
+        cachedTokens: cachedTokens2,
         ttftMs: latencyMs2,
         tpotMs: computeTpot3(latencyMs2, outputTokens2, { streaming: false })
       });
@@ -17896,6 +18055,7 @@ async function handleOpenAIResponsesProtocol(request, reply, endpoint, endpointI
         requests: 1,
         inputTokens: inputTokens2,
         outputTokens: outputTokens2,
+        cachedTokens: cachedTokens2,
         latencyMs: latencyMs2
       });
       if (storeResponsePayloads) {

package/src/web/dist/assets/{About-CH7-GnML.js → About-CfWemXks.js}

@@ -1,4 +1,4 @@
-import{c as p,u as v,a as k,r,j as e,d as o,U as j,m as i}from"./index-CMmlZ8am.js";import{u as N}from"./useApiQuery-qpXuOM9g.js";import{P as w,a as d}from"./PageSection-819eIHwY.js";import"./Input-AVaeYH0J.js";import{B as b}from"./Button-3qj8BOTb.js";import{I as y}from"./info-Cg4yNF4Z.js";/**
+import{c as p,u as v,a as k,r,j as e,d as o,U as j,m as i}from"./index-CAbsGgAq.js";import{u as N}from"./useApiQuery-CNbY7eTZ.js";import{P as w,a as d}from"./PageSection-BfMkaweN.js";import"./Input-DACzKnnk.js";import{B as b}from"./Button-DdVyALkb.js";import{I as y}from"./info-CpmSkfUl.js";/**
  * @license lucide-react v0.344.0 - ISC
  *
  * This source code is licensed under the ISC license.
@@ -8,4 +8,4 @@ import{c as p,u as v,a as k,r,j as e,d as o,U as j,m as i}from"./index-CMmlZ8am.
  *
  * This source code is licensed under the ISC license.
  * See the LICENSE file in the root directory of this source tree.
- */const E=p("Sparkles",[["path",{d:"m12 3-1.912 5.813a2 2 0 0 1-1.275 1.275L3 12l5.813 1.912a2 2 0 0 1 1.275 1.275L12 21l1.912-5.813a2 2 0 0 1 1.275-1.275L21 12l-5.813-1.912a2 2 0 0 1-1.275-1.275L12 3Z",key:"17u4zn"}],["path",{d:"M5 3v4",key:"bklmnn"}],["path",{d:"M19 17v4",key:"iiml17"}],["path",{d:"M3 5h4",key:"nem4j1"}],["path",{d:"M17 19h4",key:"lbex7p"}]]),I="0.4.0",_={version:I},L={VITE_BUILD_TIME:"2025-10-25T06:44:24.223Z",VITE_NODE_VERSION:"v22.14.0"};function m({items:t}){return t.length===0?null:e.jsx("dl",{className:"grid gap-4 sm:grid-cols-2 xl:grid-cols-2",children:t.map(s=>e.jsxs("div",{className:"rounded-2xl border border-slate-200/50 bg-white p-4 shadow-sm shadow-slate-200/30 transition-all duration-200 hover:-translate-y-0.5 hover:border-slate-200/70 hover:shadow-md hover:shadow-slate-200/40 dark:border-slate-700/50 dark:bg-slate-900/80 dark:shadow-lg dark:shadow-slate-900/30 dark:hover:border-slate-600/70",children:[e.jsx("dt",{className:"text-xs font-semibold uppercase tracking-[0.14em] text-slate-500 dark:text-slate-400",children:s.label}),e.jsx("dd",{className:"mt-2 text-base font-semibold text-slate-900 dark:text-slate-100",children:s.value}),s.hint?e.jsx("p",{className:o(i,"mt-2 text-xs leading-relaxed"),children:s.hint}):null]},s.label))})}function P(){const{t}=v(),{pushToast:s}=k(),a=N(["status","gateway"],{url:"/api/status",method:"GET"},{staleTime:6e4});r.useEffect(()=>{a.isError&&a.error&&s({title:t("about.toast.statusError.title"),description:a.error.message,variant:"error"})},[a.isError,a.error,s,t]);const n=_.version,l=r.useMemo(()=>{const u=L,f=u.VITE_BUILD_TIME,g=u.VITE_NODE_VERSION;return{buildTime:f,nodeVersion:g}},[]),h=r.useMemo(()=>[{label:t("about.app.labels.name"),value:e.jsx("span",{className:"font-mono text-sm font-semibold text-slate-900 dark:text-slate-100",children:"cc-gw"})},{label:t("about.app.labels.version"),value:e.jsxs("span",{className:"font-mono text-sm font-semibold text-blue-700 dark:text-blue-200",children:["v",n]})},{label:t("about.app.labels.buildTime"),value:l.buildTime,hint:t("about.app.hint.buildTime")},{label:t("about.app.labels.node"),value:e.jsx("span",{className:"font-mono text-sm text-slate-800 dark:text-slate-200",children:l.nodeVersion})}],[n,l.buildTime,l.nodeVersion,t]),c=r.useMemo(()=>a.data?[{label:t("about.status.labels.host"),value:a.data.host??"127.0.0.1"},{label:t("about.status.labels.port"),value:a.data.port.toLocaleString()},{label:t("about.status.labels.providers"),value:a.data.providers.toLocaleString()},{label:t("about.status.labels.active"),value:(a.data.activeRequests??0).toLocaleString(),hint:t("about.status.hint.active")}]:[],[a.data,t]),x=()=>{s({title:t("about.toast.updatesPlanned"),variant:"info"})};return e.jsxs("div",{className:"space-y-8",children:[e.jsx(w,{icon:e.jsx(y,{className:"h-6 w-6","aria-hidden":"true"}),title:t("about.title"),description:t("about.description"),badge:`v${n}`,actions:e.jsx(b,{variant:"primary",icon:e.jsx(E,{className:"h-4 w-4","aria-hidden":"true"}),onClick:x,children:t("about.support.actions.checkUpdates")})}),e.jsxs("div",{className:"grid gap-6 lg:grid-cols-2",children:[e.jsx(d,{title:t("about.app.title"),description:t("about.app.subtitle"),className:"h-full",contentClassName:"gap-4",children:e.jsx(m,{items:h})}),e.jsx(d,{title:t("about.status.title"),description:t("about.status.subtitle"),className:"h-full",contentClassName:"gap-4",actions:e.jsx(b,{variant:"subtle",size:"sm",icon:e.jsx(T,{className:"h-4 w-4","aria-hidden":"true"}),onClick:()=>a.refetch(),loading:a.isFetching,children:a.isFetching?t("common.actions.refreshing"):t("common.actions.refresh")}),children:a.isLoading?e.jsxs("div",{className:"flex h-36 flex-col items-center justify-center gap-3 text-center",children:[e.jsx("div",{className:"h-10 w-10 animate-spin rounded-full border-[3px] border-blue-500/30 border-t-blue-600 dark:border-blue-400/20 dark:border-t-blue-300"}),e.jsx("p",{className:o(i,"text-sm"),children:t("about.status.loading")})]}):c.length>0?e.jsx(m,{items:c}):e.jsxs("div",{className:"flex h-36 flex-col items-center justify-center gap-2 rounded-2xl border border-dashed border-slate-200/60 bg-white p-6 text-center shadow-inner dark:border-slate-700/60 dark:bg-slate-900/60",children:[e.jsx("p",{className:"text-sm font-semibold text-slate-700 dark:text-slate-200",children:t("about.status.empty")}),e.jsx("p",{className:o(i,"text-xs"),children:t("common.actions.refresh")})]})})]}),e.jsx(d,{title:t("about.support.title"),description:e.jsxs("span",{className:"space-y-1",children:[e.jsx("span",{className:"block text-sm font-semibold text-blue-600 dark:text-blue-300",children:t("about.support.subtitle")}),e.jsx("span",{children:t("about.support.description")})]}),className:"relative overflow-hidden",contentClassName:"gap-6",children:e.jsxs("div",{className:"flex flex-col gap-4 rounded-3xl border border-slate-200/50 bg-white p-6 shadow-lg shadow-slate-200/30 backdrop-blur-md dark:border-slate-700/50 dark:bg-slate-900/80 dark:shadow-slate-900/40",children:[e.jsxs("div",{className:"flex flex-wrap items-start gap-4",children:[e.jsx("div",{className:"grid h-12 w-12 place-items-center rounded-2xl bg-gradient-to-br from-blue-500/20 to-indigo-500/20 text-blue-600 shadow-inner dark:text-blue-200",children:e.jsx(j,{className:"h-6 w-6","aria-hidden":"true"})}),e.jsx("p",{className:o(i,"text-sm leading-6"),children:t("about.support.tip")})]}),e.jsx("code",{className:"inline-flex items-center gap-2 self-start rounded-full border border-blue-200/50 bg-blue-50/80 px-4 py-2 text-xs font-semibold tracking-wide text-blue-700 shadow-sm dark:border-blue-500/30 dark:bg-blue-900/30 dark:text-blue-200",children:"~/.cc-gw/config.json"})]})})]})}export{P as default};
+ */const E=p("Sparkles",[["path",{d:"m12 3-1.912 5.813a2 2 0 0 1-1.275 1.275L3 12l5.813 1.912a2 2 0 0 1 1.275 1.275L12 21l1.912-5.813a2 2 0 0 1 1.275-1.275L21 12l-5.813-1.912a2 2 0 0 1-1.275-1.275L12 3Z",key:"17u4zn"}],["path",{d:"M5 3v4",key:"bklmnn"}],["path",{d:"M19 17v4",key:"iiml17"}],["path",{d:"M3 5h4",key:"nem4j1"}],["path",{d:"M17 19h4",key:"lbex7p"}]]),I="0.4.2",_={version:I},L={VITE_BUILD_TIME:"2025-10-29T03:49:27.299Z",VITE_NODE_VERSION:"v22.14.0"};function m({items:t}){return t.length===0?null:e.jsx("dl",{className:"grid gap-4 sm:grid-cols-2 xl:grid-cols-2",children:t.map(s=>e.jsxs("div",{className:"rounded-2xl border border-slate-200/50 bg-white p-4 shadow-sm shadow-slate-200/30 transition-all duration-200 hover:-translate-y-0.5 hover:border-slate-200/70 hover:shadow-md hover:shadow-slate-200/40 dark:border-slate-700/50 dark:bg-slate-900/80 dark:shadow-lg dark:shadow-slate-900/30 dark:hover:border-slate-600/70",children:[e.jsx("dt",{className:"text-xs font-semibold uppercase tracking-[0.14em] text-slate-500 dark:text-slate-400",children:s.label}),e.jsx("dd",{className:"mt-2 text-base font-semibold text-slate-900 dark:text-slate-100",children:s.value}),s.hint?e.jsx("p",{className:o(i,"mt-2 text-xs leading-relaxed"),children:s.hint}):null]},s.label))})}function P(){const{t}=v(),{pushToast:s}=k(),a=N(["status","gateway"],{url:"/api/status",method:"GET"},{staleTime:6e4});r.useEffect(()=>{a.isError&&a.error&&s({title:t("about.toast.statusError.title"),description:a.error.message,variant:"error"})},[a.isError,a.error,s,t]);const n=_.version,l=r.useMemo(()=>{const u=L,f=u.VITE_BUILD_TIME,g=u.VITE_NODE_VERSION;return{buildTime:f,nodeVersion:g}},[]),h=r.useMemo(()=>[{label:t("about.app.labels.name"),value:e.jsx("span",{className:"font-mono text-sm font-semibold text-slate-900 dark:text-slate-100",children:"cc-gw"})},{label:t("about.app.labels.version"),value:e.jsxs("span",{className:"font-mono text-sm font-semibold text-blue-700 dark:text-blue-200",children:["v",n]})},{label:t("about.app.labels.buildTime"),value:l.buildTime,hint:t("about.app.hint.buildTime")},{label:t("about.app.labels.node"),value:e.jsx("span",{className:"font-mono text-sm text-slate-800 dark:text-slate-200",children:l.nodeVersion})}],[n,l.buildTime,l.nodeVersion,t]),c=r.useMemo(()=>a.data?[{label:t("about.status.labels.host"),value:a.data.host??"127.0.0.1"},{label:t("about.status.labels.port"),value:a.data.port.toLocaleString()},{label:t("about.status.labels.providers"),value:a.data.providers.toLocaleString()},{label:t("about.status.labels.active"),value:(a.data.activeRequests??0).toLocaleString(),hint:t("about.status.hint.active")}]:[],[a.data,t]),x=()=>{s({title:t("about.toast.updatesPlanned"),variant:"info"})};return e.jsxs("div",{className:"space-y-8",children:[e.jsx(w,{icon:e.jsx(y,{className:"h-6 w-6","aria-hidden":"true"}),title:t("about.title"),description:t("about.description"),badge:`v${n}`,actions:e.jsx(b,{variant:"primary",icon:e.jsx(E,{className:"h-4 w-4","aria-hidden":"true"}),onClick:x,children:t("about.support.actions.checkUpdates")})}),e.jsxs("div",{className:"grid gap-6 lg:grid-cols-2",children:[e.jsx(d,{title:t("about.app.title"),description:t("about.app.subtitle"),className:"h-full",contentClassName:"gap-4",children:e.jsx(m,{items:h})}),e.jsx(d,{title:t("about.status.title"),description:t("about.status.subtitle"),className:"h-full",contentClassName:"gap-4",actions:e.jsx(b,{variant:"subtle",size:"sm",icon:e.jsx(T,{className:"h-4 w-4","aria-hidden":"true"}),onClick:()=>a.refetch(),loading:a.isFetching,children:a.isFetching?t("common.actions.refreshing"):t("common.actions.refresh")}),children:a.isLoading?e.jsxs("div",{className:"flex h-36 flex-col items-center justify-center gap-3 text-center",children:[e.jsx("div",{className:"h-10 w-10 animate-spin rounded-full border-[3px] border-blue-500/30 border-t-blue-600 dark:border-blue-400/20 dark:border-t-blue-300"}),e.jsx("p",{className:o(i,"text-sm"),children:t("about.status.loading")})]}):c.length>0?e.jsx(m,{items:c}):e.jsxs("div",{className:"flex h-36 flex-col items-center justify-center gap-2 rounded-2xl border border-dashed border-slate-200/60 bg-white p-6 text-center shadow-inner dark:border-slate-700/60 dark:bg-slate-900/60",children:[e.jsx("p",{className:"text-sm font-semibold text-slate-700 dark:text-slate-200",children:t("about.status.empty")}),e.jsx("p",{className:o(i,"text-xs"),children:t("common.actions.refresh")})]})})]}),e.jsx(d,{title:t("about.support.title"),description:e.jsxs("span",{className:"space-y-1",children:[e.jsx("span",{className:"block text-sm font-semibold text-blue-600 dark:text-blue-300",children:t("about.support.subtitle")}),e.jsx("span",{children:t("about.support.description")})]}),className:"relative overflow-hidden",contentClassName:"gap-6",children:e.jsxs("div",{className:"flex flex-col gap-4 rounded-3xl border border-slate-200/50 bg-white p-6 shadow-lg shadow-slate-200/30 backdrop-blur-md dark:border-slate-700/50 dark:bg-slate-900/80 dark:shadow-slate-900/40",children:[e.jsxs("div",{className:"flex flex-wrap items-start gap-4",children:[e.jsx("div",{className:"grid h-12 w-12 place-items-center rounded-2xl bg-gradient-to-br from-blue-500/20 to-indigo-500/20 text-blue-600 shadow-inner dark:text-blue-200",children:e.jsx(j,{className:"h-6 w-6","aria-hidden":"true"})}),e.jsx("p",{className:o(i,"text-sm leading-6"),children:t("about.support.tip")})]}),e.jsx("code",{className:"inline-flex items-center gap-2 self-start rounded-full border border-blue-200/50 bg-blue-50/80 px-4 py-2 text-xs font-semibold tracking-wide text-blue-700 shadow-sm dark:border-blue-500/30 dark:bg-blue-900/30 dark:text-blue-200",children:"~/.cc-gw/config.json"})]})})]})}export{P as default};

package/src/web/dist/assets/{ApiKeys-DbxeLLKa.js → ApiKeys-jGq-kKf8.js}

@@ -1,4 +1,4 @@
-import{c as T,u as W,a as J,r as d,j as e,L as X,N as Y,d as t,b as K,m as r,H as m,E as Z,k as ee,l as ae,f as se,h as te}from"./index-CMmlZ8am.js";import{E as le}from"./index-D4f4GIYb.js";import{u as w}from"./useApiQuery-qpXuOM9g.js";import{P as ie,a as q}from"./PageSection-819eIHwY.js";import{F as I}from"./FormField-DmKxUoqD.js";import{I as re}from"./Input-AVaeYH0J.js";import{B as C}from"./Button-3qj8BOTb.js";import{S as ne}from"./StatusBadge-FIJj_rsA.js";import{C as ce}from"./copy-dJsIfcXG.js";/**
+import{c as T,u as W,a as J,r as d,j as e,L as X,N as Y,d as t,b as K,m as r,H as m,E as Z,k as ee,l as ae,f as se,h as te}from"./index-CAbsGgAq.js";import{E as le}from"./index-C2xoexDc.js";import{u as w}from"./useApiQuery-CNbY7eTZ.js";import{P as ie,a as q}from"./PageSection-BfMkaweN.js";import{F as I}from"./FormField-Dyi1Kr0k.js";import{I as re}from"./Input-DACzKnnk.js";import{B as C}from"./Button-DdVyALkb.js";import{S as ne}from"./StatusBadge-OH4R_aKr.js";import{C as ce}from"./copy-BVgnqUsP.js";/**
  * @license lucide-react v0.344.0 - ISC
  *
  * This source code is licensed under the ISC license.

package/src/web/dist/assets/{Button-3qj8BOTb.js → Button-DdVyALkb.js}

@@ -1 +1 @@
-import{j as s,d as c,J as i,E as p,k as b}from"./index-CMmlZ8am.js";function h({variant:n="subtle",size:o="md",children:a,loading:t,icon:e,className:l,disabled:r,...u}){const m={subtle:b,primary:p,danger:i}[n],x={sm:"h-8 px-3 text-xs",md:"h-10 px-4 text-sm",lg:"h-12 px-6 text-base"}[o];return s.jsx("button",{className:c(m,x,t&&"cursor-wait opacity-70",r&&"cursor-not-allowed opacity-50",l),disabled:r||t,...u,children:t?s.jsx("div",{className:"inline-block animate-spin rounded-full h-4 w-4 border-b-2 border-current"}):e?s.jsxs(s.Fragment,{children:[e,a]}):a})}export{h as B};
+import{j as s,d as c,J as i,E as p,k as b}from"./index-CAbsGgAq.js";function h({variant:n="subtle",size:o="md",children:a,loading:t,icon:e,className:l,disabled:r,...u}){const m={subtle:b,primary:p,danger:i}[n],x={sm:"h-8 px-3 text-xs",md:"h-10 px-4 text-sm",lg:"h-12 px-6 text-base"}[o];return s.jsx("button",{className:c(m,x,t&&"cursor-wait opacity-70",r&&"cursor-not-allowed opacity-50",l),disabled:r||t,...u,children:t?s.jsx("div",{className:"inline-block animate-spin rounded-full h-4 w-4 border-b-2 border-current"}):e?s.jsxs(s.Fragment,{children:[e,a]}):a})}export{h as B};