From 0eff634bd063ec084eb6a8deb3f9fecfe73ad51e Mon Sep 17 00:00:00 2001
From: Cohee <18619528+Cohee1207@users.noreply.github.com>
Date: Wed, 4 Jun 2025 20:30:20 +0300
Subject: [PATCH] OpenRouter: add cache TTL control for Claude

---
 src/endpoints/backends/chat-completions.js | 5 +++--
 src/prompt-converters.js                   | 6 ++++--
 2 files changed, 7 insertions(+), 4 deletions(-)

diff --git a/src/endpoints/backends/chat-completions.js b/src/endpoints/backends/chat-completions.js
index c8a7c5f7e..ba2c28129 100644
--- a/src/endpoints/backends/chat-completions.js
+++ b/src/endpoints/backends/chat-completions.js
@@ -1342,10 +1342,11 @@ router.post('/generate', function (request, response) {
             bodyParams['reasoning'] = { effort: request.body.reasoning_effort };
         }
 
-        let cachingAtDepth = getConfigValue('claude.cachingAtDepth', -1, 'number');
+        const cachingAtDepth = getConfigValue('claude.cachingAtDepth', -1, 'number');
         const isClaude3or4 = /anthropic\/claude-(3|opus-4|sonnet-4)/.test(request.body.model);
+        const cacheTTL = getConfigValue('claude.extendedTTL', false, 'boolean') ? '1h' : '5m';
         if (Number.isInteger(cachingAtDepth) && cachingAtDepth >= 0 && isClaude3or4) {
-            cachingAtDepthForOpenRouterClaude(request.body.messages, cachingAtDepth);
+            cachingAtDepthForOpenRouterClaude(request.body.messages, cachingAtDepth, cacheTTL);
         }
     } else if (request.body.chat_completion_source === CHAT_COMPLETION_SOURCES.CUSTOM) {
         apiUrl = request.body.custom_url;
diff --git a/src/prompt-converters.js b/src/prompt-converters.js
index eab185c7d..0ce1cc3ed 100644
--- a/src/prompt-converters.js
+++ b/src/prompt-converters.js
@@ -906,8 +906,9 @@ export function cachingAtDepthForClaude(messages, cachingAtDepth, ttl) {
  * messages array.
  * @param {object[]} messages Array of messages
  * @param {number} cachingAtDepth Depth at which caching is supposed to occur
+ * @param {string} ttl TTL value
  */
-export function cachingAtDepthForOpenRouterClaude(messages, cachingAtDepth) {
+export function cachingAtDepthForOpenRouterClaude(messages, cachingAtDepth, ttl) {
     //caching the prefill is a terrible idea in general
     let passedThePrefill = false;
     //depth here is the number of message role switches
@@ -927,12 +928,13 @@ export function cachingAtDepthForOpenRouterClaude(messages, cachingAtDepth) {
                     messages[i].content = [{
                         type: 'text',
                         text: content,
-                        cache_control: { type: 'ephemeral' },
+                        cache_control: { type: 'ephemeral', ttl: ttl },
                     }];
                 } else {
                     const contentPartCount = content.length;
                     content[contentPartCount - 1].cache_control = {
                         type: 'ephemeral',
+                        ttl: ttl,
                     };
                 }
             }