Claude: set cache TTL to 1h

22afd570fcf659e40b66fb3803821edf2e550842

Cohee <18619528+Cohee1207@users.noreply.github.com>

src/endpoints/backends/chat-completions.js+4 -3 Viewed

         };
         if (useSystemPrompt) {
             if (enableSystemPromptCache && Array.isArray(convertedPrompt.systemPrompt) && convertedPrompt.systemPrompt.length) {
-                convertedPrompt.systemPrompt[convertedPrompt.systemPrompt.length - 1]['cache_control'] = { type: 'ephemeral' };
+                convertedPrompt.systemPrompt[convertedPrompt.systemPrompt.length - 1]['cache_control'] = { type: 'ephemeral', ttl: '1h' };
+            }
             requestBody.system = convertedPrompt.systemPrompt;

@@ -190,7 +190,7 @@ async function sendClaudeRequest(request, response) {
190	.map(fn => ({ name: fn.name, description: fn.description, input_schema: fn.parameters }));	190	.map(fn => ({ name: fn.name, description: fn.description, input_schema: fn.parameters }));
191		191
192	if (enableSystemPromptCache && requestBody.tools.length) {	192	if (enableSystemPromptCache && requestBody.tools.length) {
193	requestBody.tools[requestBody.tools.length - 1]['cache_control'] = { type: 'ephemeral' };	193	requestBody.tools[requestBody.tools.length - 1]['cache_control'] = { type: 'ephemeral', ttl: '1h' };
194	}	194	}
195	}	195	}
196		196

                 'type': 'web_search_20250305',
                 'name': 'web_search',
             }];
-            requestBody.tools = [...(requestBody.tools || []), ...webSearchTool];
+            requestBody.tools = [...webSearchTool, ...(requestBody.tools || [])];
+        }
         if (cachingAtDepth !== -1) {

         if (enableSystemPromptCache || cachingAtDepth !== -1) {
             betaHeaders.push('prompt-caching-2024-07-31');
+            betaHeaders.push('extended-cache-ttl-2025-04-11');
+        }
         const reasoningEffort = request.body.reasoning_effort;

src/prompt-converters.js+1 -1 Viewed

         if (messages[i].role !== previousRoleName) {
             if (depth === cachingAtDepth || depth === cachingAtDepth + 2) {
                 const content = messages[i].content;
-                content[content.length - 1].cache_control = { type: 'ephemeral' };
+                content[content.length - 1].cache_control = { type: 'ephemeral', ttl: '1h' };
+            }
             if (depth === cachingAtDepth + 2) {