@absolutejs/ai 0.0.26 → 0.0.28

package/dist/ai/index.js

@@ -571,10 +571,15 @@ var flushPendingToolCalls = function* (pendingToolCalls) {
   }
   pendingToolCalls.clear();
 };
-var extractUsage = (parsedUsage) => ({
-  inputTokens: parsedUsage.prompt_tokens ?? 0,
-  outputTokens: parsedUsage.completion_tokens ?? 0
-});
+var extractUsage = (parsedUsage) => {
+  const prompt = parsedUsage.prompt_tokens ?? 0;
+  const cached = parsedUsage.cached_tokens ?? 0;
+  return {
+    cacheReadInputTokens: cached,
+    inputTokens: Math.max(0, prompt - cached),
+    outputTokens: parsedUsage.completion_tokens ?? 0
+  };
+};
 var resolveToolCallIndex = (toolCall) => {
   const raw = typeof toolCall.index === "number" ? toolCall.index : NOT_FOUND;
   return raw < 0 ? undefined : raw;
@@ -644,7 +649,9 @@ var narrowUsageRecord = (parsed) => {
   const { usage } = parsed;
   const promptTokens = typeof usage.prompt_tokens === "number" ? usage.prompt_tokens : 0;
   const completionTokens = typeof usage.completion_tokens === "number" ? usage.completion_tokens : 0;
+  const cachedTokens = isRecord(usage.prompt_tokens_details) && typeof usage.prompt_tokens_details.cached_tokens === "number" ? usage.prompt_tokens_details.cached_tokens : 0;
   return extractUsage({
+    cached_tokens: cachedTokens,
     completion_tokens: completionTokens,
     prompt_tokens: promptTokens
   });
@@ -978,8 +985,11 @@ var extractUsage2 = (response) => {
     return;
   }
   const { usage } = response;
+  const input = typeof usage.input_tokens === "number" ? usage.input_tokens : 0;
+  const cached = isRecord2(usage.input_tokens_details) && typeof usage.input_tokens_details.cached_tokens === "number" ? usage.input_tokens_details.cached_tokens : 0;
   return {
-    inputTokens: typeof usage.input_tokens === "number" ? usage.input_tokens : 0,
+    cacheReadInputTokens: cached,
+    inputTokens: Math.max(0, input - cached),
     outputTokens: typeof usage.output_tokens === "number" ? usage.output_tokens : 0
   };
 };
@@ -1591,9 +1601,11 @@ var cacheLastContentBlock = (msg) => {
   };
   return { content: blocks, role: msg.role };
 };
-var buildRequestBody4 = (params, configuredMax, promptCaching) => {
+var buildRequestBody4 = (params, configuredMax, configCaching) => {
+  const caching = params.promptCaching ?? configCaching;
+  const cacheSystem = params.cacheSystemPrompt ?? caching;
   const messages = params.messages.filter((msg) => msg.role !== "system").map(mapMessage);
-  if (promptCaching && messages.length > 1) {
+  if (caching && messages.length > 1) {
     const last = messages[messages.length - 1];
     if (last) {
       messages[messages.length - 1] = cacheLastContentBlock(last);
@@ -1607,7 +1619,7 @@ var buildRequestBody4 = (params, configuredMax, promptCaching) => {
     stream: true
   };
   if (params.systemPrompt) {
-    body.system = promptCaching || params.cacheSystemPrompt ? [
+    body.system = cacheSystem ? [
       {
         cache_control: { type: "ephemeral" },
         text: params.systemPrompt,
@@ -1617,7 +1629,7 @@ var buildRequestBody4 = (params, configuredMax, promptCaching) => {
   }
   if (params.tools && params.tools.length > 0) {
     const tools = params.tools.map(mapToolDefinition2);
-    if (promptCaching) {
+    if (caching) {
       tools[tools.length - 1] = {
         ...tools[tools.length - 1],
         cache_control: { type: "ephemeral" }
@@ -2511,8 +2523,10 @@ var processToolTurn = async (socket, options, state, messageId, conversationId,
   });
   const toolDefs = options.tools ? buildToolDefinitions(options.tools) : undefined;
   const stream = options.provider.stream({
+    cacheSystemPrompt: options.cacheSystemPrompt,
     messages: state.currentMessages,
     model: options.model,
+    promptCaching: options.promptCaching,
     reasoning: options.reasoning,
     signal,
     systemPrompt: options.systemPrompt,
@@ -2595,8 +2609,10 @@ var processStreamTextChunk = async (chunk, options, socket, messageId, conversat
 var processStream = async (socket, options, messages, messageId, conversationId, signal, startTime) => {
   const toolDefs = options.tools ? buildToolDefinitions(options.tools) : undefined;
   const stream = options.provider.stream({
+    cacheSystemPrompt: options.cacheSystemPrompt,
     messages,
     model: options.model,
+    promptCaching: options.promptCaching,
     reasoning: options.reasoning,
     signal,
     systemPrompt: options.systemPrompt,
@@ -2944,9 +2960,11 @@ var streamTurns = async function* (options, renderers, messages, signal, startTi
       usage: undefined
     };
     const stream = options.provider.stream({
+      cacheSystemPrompt: options.cacheSystemPrompt,
       maxTokens: options.maxTokens,
       messages: turnState.currentMessages,
       model: options.model,
+      promptCaching: options.promptCaching,
       reasoning: options.reasoning,
       signal,
       systemPrompt: options.systemPrompt,
@@ -3321,6 +3339,7 @@ var generateAI = async (options) => {
     maxTokens: options.maxTokens,
     messages: options.messages,
     model: options.model,
+    promptCaching: options.promptCaching,
     reasoning: options.reasoning,
     responseFormat: options.responseFormat,
     signal: options.signal,
@@ -3425,6 +3444,7 @@ var generateObjectAI = async (options) => {
     maxTokens: options.maxTokens,
     messages: options.messages,
     model: options.model,
+    promptCaching: options.promptCaching,
     provider: options.provider,
     reasoning: options.reasoning,
     signal: options.signal,
@@ -4400,5 +4420,5 @@ export {
   PROVIDER_STATUS_PAGES
 };
 
-//# debugId=90792F1972AEBAA464756E2164756E21
+//# debugId=CE76A2131808230464756E2164756E21
 //# sourceMappingURL=index.js.map