Add configurable Anthropic prompt cache TTL settings

src/renderer/modals/SessionSettings.tsx

@@ -568,7 +568,33 @@ export function ChatConfig({
           </Flex>
         </Stack>
       )}
+      {settings?.provider === ModelProviderEnum.Claude && (
+        <Stack gap="xs">
+          <Text size="sm" fw="600">
+            {t('Prompt Cache')}
+          </Text>
 
+          <SegmentedControl
+            value={settings?.providerOptions?.claude?.cacheTTL ?? 'auto'}
+            onChange={(value) =>
+              onSettingsChange({
+                providerOptions: {
+                  ...settings?.providerOptions,
+                  claude: {
+                    ...settings?.providerOptions?.claude,
+                    cacheTTL: value as 'auto' | '5m' | '1h',
+                  },
+                },
+              })
+            }
+            data={[
+              { label: t('Auto'), value: 'auto' },
+              { label: t('5 min'), value: '5m' },
+              { label: t('1 hour'), value: '1h' },
+            ]}
+          />
+        </Stack>
+      )}
       <ReasoningControlConfig settings={settings} onSettingsChange={onSettingsChange} />
     </Stack>
   )

src/shared/providers/definitions/claude.ts

@@ -77,6 +77,7 @@ export const claudeProvider = defineProvider({
         topP: config.settings.topP,
         maxOutputTokens: config.settings.maxTokens,
         stream: config.settings.stream,
+        cacheTTL: config.settings.providerOptions?.claude?.cacheTTL,
         extraHeaders: oauthHeaders,
         customFetch:
           isOAuth && credentialManager ? createBearerOAuthFetch(config.dependencies, credentialManager) : undefined,

src/shared/providers/definitions/models/claude.ts

@@ -17,6 +17,7 @@ interface Options {
   topP?: number
   maxOutputTokens?: number
   stream?: boolean
+  cacheTTL?: 'auto' | '5m' | '1h'
   extraHeaders?: Record<string, string>
   customFetch?: typeof globalThis.fetch
   authToken?: string
@@ -132,14 +133,30 @@ export default class Claude extends AbstractAISDKModel {
   }
 
   public async chat(messages: ModelMessage[], options: CallChatCompletionOptions): Promise<StreamTextResult> {
-    return super.chat(addAnthropicCacheControl(messages), options)
+    const ttl =
+  this.options.cacheTTL === '5m'
+    ? '5m'
+    : this.options.cacheTTL === '1h'
+      ? '1h'
+      : messages.length >= 10
+        ? '1h'
+        : '5m'
+    return super.chat(addAnthropicCacheControl(messages, ttl), options)
   }
 
   public async *chatStream<T extends ToolSet>(
     messages: ModelMessage[],
     options: ChatStreamOptions
   ): AsyncGenerator<ModelStreamPart<T>> {
-    yield* super.chatStream<T>(addAnthropicCacheControl(messages), options)
+    const ttl =
+  this.options.cacheTTL === '5m'
+    ? '5m'
+    : this.options.cacheTTL === '1h'
+      ? '1h'
+      : messages.length >= 10
+        ? '1h'
+        : '5m'
+    yield* super.chatStream<T>(addAnthropicCacheControl(messages, ttl), options)
   }
 
   // https://docs.anthropic.com/en/docs/api/models

src/shared/types/settings.ts

@@ -117,6 +117,8 @@ const ProviderBaseInfoSchema = z.discriminatedUnion('isCustom', [
   CustomProviderBaseInfoSchema,
 ])
 
+export const ClaudeCacheTTLSchema = z.enum(['auto', '5m', '1h'])
+
 const ClaudeParamsSchema = z.object({
   thinking: z
     .object({
@@ -125,7 +127,9 @@ const ClaudeParamsSchema = z.object({
     })
     .optional()
     .catch(undefined),
+
   effort: z.enum(['low', 'medium', 'high']).optional().catch(undefined),
+  cacheTTL: ClaudeCacheTTLSchema.optional().catch('auto'),
 })
 
 const OpenAIParamsSchema = z.object({