@codeproxy/core 0.1.4 → 0.1.6

package/README.md

@@ -23,7 +23,7 @@ const client = new OpenAI({
 });
 
 const response = await client.responses.create({
-  model: 'deepseek-v4-pro',
+  model: 'deepseek-v4-flash',
   input: 'Hello!',
 });
 ```

package/README.zh-CN.md

@@ -23,7 +23,7 @@ const client = new OpenAI({
 });
 
 const response = await client.responses.create({
-  model: 'deepseek-v4-pro',
+  model: 'deepseek-v4-flash',
   input: 'Hello!',
 });
 ```

package/dist/index.cjs

@@ -45,13 +45,21 @@ var DEFAULT_REASONING_BUDGETS = {
 function translateRequest(data, options = {}) {
   const model = data.model;
   const maxTokens = typeof data.max_output_tokens === "number" && data.max_output_tokens || typeof data.max_tokens === "number" && data.max_tokens || options.defaultMaxTokens || 8192;
-  const systemBlocks = extractSystemBlocks(data.instructions);
+  let systemBlocks = extractSystemBlocks(data.instructions);
   const built = buildMessages(data, systemBlocks);
   let messages = built.messages;
-  const hasPromptCache = built.hasPromptCache;
+  let hasPromptCache = built.hasPromptCache;
+  if (data.prompt_cache_key) {
+    hasPromptCache = true;
+    systemBlocks = markBlocksForCache(systemBlocks);
+    messages = markCacheBreakpoint(messages);
+  }
   messages = repairToolAdjacency(messages);
   messages = sanitizeMessages(messages);
   messages = ensureEndsWithUser(messages);
+  if (data.prompt_cache_key) {
+    messages = markCacheBreakpoint(messages);
+  }
   const request = {
     model,
     messages,
@@ -192,7 +200,15 @@ function buildMessages(data, systemBlocks) {
           } else if (part && typeof part === "object") {
             const contentPart = part;
             if (contentPart.type === "input_text" || contentPart.type === "text" || contentPart.type === "output_text") {
-              contentBlocks.push({ type: "text", text: String(contentPart.text ?? "") });
+              const textBlock = {
+                type: "text",
+                text: String(contentPart.text ?? "")
+              };
+              const cc = contentPart.cache_control;
+              if (cc) {
+                textBlock.cache_control = cc;
+              }
+              contentBlocks.push(textBlock);
             } else if (contentPart.type === "input_image" || contentPart.type === "image" || contentPart.type === "image_url") {
               const imgUrlPart = contentPart;
               const imgUrl = imgUrlPart.image_url;
@@ -510,6 +526,43 @@ function ensureEndsWithUser(messages) {
   }
   return [...messages, { role: "user", content: [{ type: "text", text: "Continue." }] }];
 }
+function markBlocksForCache(blocks) {
+  for (const block of blocks) {
+    if (!block.cache_control) {
+      block.cache_control = { type: "ephemeral" };
+    }
+  }
+  return blocks;
+}
+function markCacheBreakpoint(messages) {
+  for (const msg of messages) {
+    if (msg.role === "assistant" && Array.isArray(msg.content)) {
+      for (let j = msg.content.length - 1; j >= 0; j--) {
+        const block = msg.content[j];
+        if (block.type === "text") {
+          if (!block.cache_control) {
+            block.cache_control = { type: "ephemeral" };
+          }
+          return messages;
+        }
+      }
+    }
+  }
+  for (const msg of messages) {
+    if (msg.role === "user" && Array.isArray(msg.content)) {
+      for (let j = msg.content.length - 1; j >= 0; j--) {
+        const block = msg.content[j];
+        if (block.type === "text") {
+          if (!block.cache_control) {
+            block.cache_control = { type: "ephemeral" };
+          }
+          return messages;
+        }
+      }
+    }
+  }
+  return messages;
+}
 
 // src/utils/json.ts
 function safeJsonParse(text) {
@@ -1961,6 +2014,12 @@ function buildUpstreamBody(request, format, streaming, baseUrl, dropImages, reas
         ar.thinking = { type: "enabled", budget_tokens: 65536 };
       }
     }
+    if (ar.thinking && typeof ar.thinking === "object" && "budget_tokens" in ar.thinking) {
+      const budget = Number(ar.thinking.budget_tokens);
+      if (budget > 0 && ar.max_tokens <= budget) {
+        ar.max_tokens = budget + 1024;
+      }
+    }
     return {
       upstreamBody: ar,
       requestMetadata: buildRequestMetadata(request, ar.temperature, ar.top_p)