feat(cache): add OpenAI prompt_cache_key and Gemini cachedContent support

Provider-specific caching enhancements:

OpenAI:
- Auto-generate prompt_cache_key from message prefix hash
- Key format: omni-{prefix_hash_32chars}
- Preserves client-provided keys (doesn't override)

Gemini:
- Preserve cachedContent ID if provided by client
- Enables explicit Gemini caching for long prompts
- cachedContentTokenCount already tracked in responses

open-sse/handlers/chatCore.ts

@@ -950,11 +950,24 @@ export async function handleChatCore({
 
   const executeProviderRequest = async (modelToCall = effectiveModel, allowDedup = false) => {
     const execute = async () => {
-      const bodyToSend =
+      let bodyToSend =
         translatedBody.model === modelToCall
           ? translatedBody
           : { ...translatedBody, model: modelToCall };
 
+      // Inject prompt_cache_key for OpenAI providers if not already set
+      if (
+        targetFormat === FORMATS.OPENAI &&
+        !bodyToSend.prompt_cache_key &&
+        Array.isArray(bodyToSend.messages)
+      ) {
+        const { generatePromptCacheKey } = await import("@/lib/promptCache");
+        const cacheKey = generatePromptCacheKey(bodyToSend.messages);
+        if (cacheKey) {
+          bodyToSend = { ...bodyToSend, prompt_cache_key: cacheKey };
+        }
+      }
+
       const rawResult = await withRateLimit(provider, connectionId, modelToCall, () =>
         executor.execute({
           model: modelToCall,
@@ -1444,11 +1457,19 @@ export async function handleChatCore({
       const cachedTokens = toPositiveNumber(
         usage.cache_read_input_tokens ??
           usage.cached_tokens ??
-          ((usage as Record<string, unknown>).prompt_tokens_details as Record<string, unknown> | undefined)?.cached_tokens
+          (
+            (usage as Record<string, unknown>).prompt_tokens_details as
+              | Record<string, unknown>
+              | undefined
+          )?.cached_tokens
       );
       const cacheCreationTokens = toPositiveNumber(
         usage.cache_creation_input_tokens ??
-          ((usage as Record<string, unknown>).prompt_tokens_details as Record<string, unknown> | undefined)?.cache_creation_tokens
+          (
+            (usage as Record<string, unknown>).prompt_tokens_details as
+              | Record<string, unknown>
+              | undefined
+          )?.cache_creation_tokens
       );
 
       saveRequestUsage({
@@ -1604,11 +1625,19 @@ export async function handleChatCore({
       const cachedTokens = toPositiveNumber(
         streamUsage.cache_read_input_tokens ??
           streamUsage.cached_tokens ??
-          ((streamUsage as Record<string, unknown>).prompt_tokens_details as Record<string, unknown> | undefined)?.cached_tokens
+          (
+            (streamUsage as Record<string, unknown>).prompt_tokens_details as
+              | Record<string, unknown>
+              | undefined
+          )?.cached_tokens
       );
       const cacheCreationTokens = toPositiveNumber(
         streamUsage.cache_creation_input_tokens ??
-          ((streamUsage as Record<string, unknown>).prompt_tokens_details as Record<string, unknown> | undefined)?.cache_creation_tokens
+          (
+            (streamUsage as Record<string, unknown>).prompt_tokens_details as
+              | Record<string, unknown>
+              | undefined
+          )?.cache_creation_tokens
       );
 
       saveRequestUsage({

open-sse/translator/request/openai-to-gemini.ts

@@ -52,6 +52,7 @@ type GeminiRequest = {
   safetySettings: unknown;
   systemInstruction?: GeminiContent;
   tools?: Array<{ functionDeclarations: GeminiFunctionDeclaration[] }>;
+  cachedContent?: string;
 };
 
 type CloudCodeEnvelope = {
@@ -82,6 +83,11 @@ function openaiToGeminiBase(model, body, stream) {
     safetySettings: DEFAULT_SAFETY_SETTINGS,
   };
 
+  // Preserve cachedContent if provided by client (for explicit Gemini caching)
+  if (body.cachedContent) {
+    result.cachedContent = body.cachedContent;
+  }
+
   // Generation config
   if (body.temperature !== undefined) {
     result.generationConfig.temperature = body.temperature;

src/lib/promptCache/index.ts

@@ -1 +1 @@
-export { analyzePrefix, shouldInjectCacheControl } from "./prefixAnalyzer";
+export { analyzePrefix, shouldInjectCacheControl, generatePromptCacheKey } from "./prefixAnalyzer";

src/lib/promptCache/prefixAnalyzer.ts

@@ -75,3 +75,11 @@ export function analyzePrefix(messages: Message[]): PrefixAnalysis {
 export function shouldInjectCacheControl(analysis: PrefixAnalysis, minTokens = 1024): boolean {
   return analysis.prefixTokens >= minTokens && analysis.confidence >= 0.7;
 }
+
+export function generatePromptCacheKey(messages: Message[]): string {
+  const analysis = analyzePrefix(messages);
+  if (analysis.prefixHash) {
+    return `omni-${analysis.prefixHash.slice(0, 32)}`;
+  }
+  return "";
+}