@bman654/clodex 2.1.3 → 2.1.4

package/dist/cli.js

@@ -216,7 +216,7 @@ import { join } from "path";
 // package.json
 var package_default = {
   name: "@bman654/clodex",
-  version: "2.1.3",
+  version: "2.1.4",
   publishConfig: {
     access: "public"
   },
@@ -9739,7 +9739,7 @@ async function writeAnthropicStream(stream, modelId, write, log12, observer, too
   let openToolId = null;
   let finishReason = "end_turn";
   let usage = {
-    input_tokens: 0,
+    input_tokens: observer?.initialInputTokens ?? 0,
     output_tokens: 0,
     cache_creation_input_tokens: 0,
     cache_read_input_tokens: 0
@@ -9758,7 +9758,7 @@ async function writeAnthropicStream(stream, modelId, write, log12, observer, too
         stop_reason: null,
         stop_sequence: null,
         usage: {
-          input_tokens: observer?.initialInputTokens ?? 0,
+          input_tokens: 0,
           output_tokens: 0,
           cache_creation_input_tokens: 0,
           cache_read_input_tokens: 0
@@ -9895,7 +9895,9 @@ async function writeAnthropicStream(stream, modelId, write, log12, observer, too
       }
       case "finish":
         if (part.totalUsage) {
-          usage = toAnthropicUsage(part.totalUsage);
+          const finalUsage = toAnthropicUsage(part.totalUsage);
+          const hasFinalInputUsage = finalUsage.input_tokens + finalUsage.cache_creation_input_tokens + finalUsage.cache_read_input_tokens > 0;
+          usage = hasFinalInputUsage ? finalUsage : { ...usage, output_tokens: finalUsage.output_tokens };
         }
         if (part.finishReason === "tool-calls") finishReason = "tool_use";
         else if (part.finishReason === "length") finishReason = "max_tokens";
@@ -10125,6 +10127,7 @@ function anthropicPromptTooLongMessage(body, contextWindow) {
 // src/proxy.ts
 var STREAM_KEEPALIVE_INTERVAL_MS = 2e4;
 var STREAM_KEEPALIVE_PING = 'event: ping\ndata: {"type":"ping"}\n\n';
+var INTERNAL_ADAPTER_KEEPALIVE_TIMEOUT_MS = 6e4;
 function createTranslationLifecycle(logPath, requestId, claudeSessionId, modelId, provider) {
   if (!logPath || !requestId) return void 0;
   const startedAt = Date.now();
@@ -10656,6 +10659,7 @@ data: ${JSON.stringify({
     }
     anthropicError(res, 404, `Unknown endpoint: ${req.method} ${req.url}`);
   });
+  server.keepAliveTimeout = INTERNAL_ADAPTER_KEEPALIVE_TIMEOUT_MS;
   let address;
   try {
     address = await listenTcpServer(server, 0, "127.0.0.1");
@@ -12152,7 +12156,7 @@ function forwardRawAnthropicRequest(req, res, rawBody, origin, rejectUnauthorize
     upstream.end(rawBody);
   });
 }
-function forwardToAdapter(req, res, rawBody, adapter, adapterRequest = http.request, lifecycle, isLocalShutdown = () => false) {
+function forwardToAdapter(req, res, rawBody, adapter, adapterRequest = http.request, adapterAgent, lifecycle, isLocalShutdown = () => false) {
   return new Promise((resolve3) => {
     const startedAt = Date.now();
     let lastActivityAt = startedAt;
@@ -12260,6 +12264,7 @@ function forwardToAdapter(req, res, rawBody, adapter, adapterRequest = http.requ
       port: adapter.port,
       method: "POST",
       path: req.url,
+      agent: adapterAgent,
       headers: {
         "Content-Type": "application/json",
         "Content-Length": String(rawBody.length),
@@ -12397,6 +12402,7 @@ async function startHttpProxy(options) {
       options.modelAliases
     );
   }
+  const adapterAgent = adapter ? new http.Agent({ keepAlive: true }) : void 0;
   let shuttingDown = false;
   const mitmServer = https.createServer({
     key: certificates.serverKey,
@@ -12490,6 +12496,7 @@ async function startHttpProxy(options) {
           adapterBody,
           adapter,
           options.adapterRequest,
+          adapterAgent,
           messagesEndpoint === "messages" && options.inferenceLogPath ? {
             logPath: options.inferenceLogPath,
             requestId,
@@ -12597,6 +12604,7 @@ async function startHttpProxy(options) {
       options.host ?? "127.0.0.1"
     );
   } catch (err) {
+    adapterAgent?.destroy();
     adapter?.close();
     throw err;
   }