[Electron Hub] Prompt Caching Support for Claude models (#4918) * Prompt Caching support Claude models * Prompt Caching support Claude models * Diff clean-up --------- Co-authored-by: Cohee <18619528+Cohee1207@users.noreply.github.com>

829db7f2d059daa5e9b95696e3029e87d4ba3ddf

Ngo Dinh Gia Bao <112873281+snowby666@users.noreply.github.com>

Signed

src/endpoints/backends/chat-completions.js+28 -18 Viewed

 const API_OPENROUTER = 'https://openrouter.ai/api/v1';
 /**
+ * Module-scoped Claude caching configuration values.
+ */
+const cacheTTL = getConfigValue('claude.extendedTTL', false, 'boolean') ? '1h' : '5m';
+const enableSystemPromptCache = getConfigValue('claude.enableSystemPromptCache', false, 'boolean');
+const cachingAtDepth = (() => {
+    const value = getConfigValue('claude.cachingAtDepth', -1, 'number');
+    return Number.isInteger(value) && value >= 0 ? value : -1;
+})();
+/**
  * Cache for cacheable (writing) OpenRouter model IDs.
  * @type {string[]}
  */

     const apiUrl = new URL(request.body.reverse_proxy || API_CLAUDE).toString();
     const apiKey = request.body.reverse_proxy ? request.body.proxy_password : readSecret(request.user.directories, SECRET_KEYS.CLAUDE);
     const divider = '-'.repeat(process.stdout.columns);
-    const enableSystemPromptCache = getConfigValue('claude.enableSystemPromptCache', false, 'boolean');
-    let cachingAtDepth = getConfigValue('claude.cachingAtDepth', -1, 'number');
-    // Disabled if not an integer or negative
-    if (!Number.isInteger(cachingAtDepth) || cachingAtDepth < 0) {
-        cachingAtDepth = -1;
-    }
     if (!apiKey) {
         console.warn(color.red(`Claude API key is missing.\n${divider}`));

         const useWebSearch = /^claude-(3-5|3-7|opus-4|sonnet-4|haiku-4-5|opus-4-5)/.test(request.body.model) && Boolean(request.body.enable_web_search);
         const isLimitedSampling = /^claude-(opus-4-1|sonnet-4-5|haiku-4-5|opus-4-5)/.test(request.body.model);
         const useVerbosity = /^claude-(opus-4-5)/.test(request.body.model);
-        const cacheTTL = getConfigValue('claude.extendedTTL', false, 'boolean') ? '1h' : '5m';
         let fixThinkingPrefill = false;
         // Add custom stop sequences
         const stopSequences = [];

             };
+        }
+        const isClaude = /^claude-/.test(request.body.model);
+        if (Array.isArray(request.body.messages) && isClaude) {
+            if (enableSystemPromptCache) {
+                cachingSystemPromptForOpenRouter(request.body.messages, cacheTTL);
+            }
+            if (cachingAtDepth !== -1) {
+                cachingAtDepthForOpenRouterClaude(request.body.messages, cachingAtDepth, cacheTTL);
+            }
+        }
         const requestBody = {
             'messages': request.body.messages,
             'model': request.body.model,

                 };
+            }
-            const enableSystemPromptCache = getConfigValue('claude.enableSystemPromptCache', false, 'boolean');
+            const isClaude = /^anthropic\/claude/.test(request.body.model);
-            const cachingAtDepth = getConfigValue('claude.cachingAtDepth', -1, 'number');
-            const isClaude3or4 = /anthropic\/claude-(3|opus-4|sonnet-4|haiku-4)/.test(request.body.model);
-            const cacheTTL = getConfigValue('claude.extendedTTL', false, 'boolean') ? '1h' : '5m';
             const isGemini = /google\/gemini/.test(request.body.model);
             const isCacheableGemini = isGemini && await isOpenRouterModelCacheable(request.body.model);
             const enableGeminiSystemPromptCache = getConfigValue('gemini.enableSystemPromptCache', false, 'boolean');

@@ -2114,12 +2125,12 @@ router.post('/generate', async function (request, response) {
2114	embedOpenRouterMedia(request.body.messages);	2125	embedOpenRouterMedia(request.body.messages);
2115	addOpenRouterSignatures(request.body.messages, request.body.model);	2126	addOpenRouterSignatures(request.body.messages, request.body.model);
2116		2127
2117	if (isClaude3or4) {	2128	if (isClaude) {
2118	if (enableSystemPromptCache) {	2129	if (enableSystemPromptCache) {
2119	cachingSystemPromptForOpenRouter(request.body.messages, cacheTTL);	2130	cachingSystemPromptForOpenRouter(request.body.messages, cacheTTL);
2120	}	2131	}
2121		2132
2122	if (Number.isInteger(cachingAtDepth) && cachingAtDepth >= 0) {	2133	if (cachingAtDepth !== -1) {
2123	cachingAtDepthForOpenRouterClaude(request.body.messages, cachingAtDepth, cacheTTL);	2134	cachingAtDepthForOpenRouterClaude(request.body.messages, cachingAtDepth, cacheTTL);
2124	}	2135	}
2125	}	2136	}

             if (request.body.repetition_penalty !== undefined) {
                 bodyParams['repetition_penalty'] = request.body.repetition_penalty;
+            }
-            const enableSystemPromptCache = getConfigValue('claude.enableSystemPromptCache', false, 'boolean');
-            const isClaude3or4 = /claude-(3|opus-4|sonnet-4|haiku-4)/.test(request.body.model);
+            const isClaude = /^claude-/.test(request.body.model);
-            const cacheTTL = getConfigValue('claude.extendedTTL', false, 'boolean') ? '1h' : '5m';
+            if (enableSystemPromptCache && isClaude) {
-            if (enableSystemPromptCache && isClaude3or4) {
                 bodyParams['cache_control'] = {
                     'enabled': true,
                     'ttl': cacheTTL,