@posthog/ai 8.7.0 → 8.7.1

package/dist/vercel/index.mjs

@@ -172,11 +172,16 @@ function redactBase64DataUrl(str, mediaType) {
 const sanitizeVercel = data => redactor.redact(data);
 
 const TOKEN_PROPERTY_KEYS = new Set(['$ai_input_tokens', '$ai_output_tokens', '$ai_cache_read_input_tokens', '$ai_cache_creation_input_tokens', '$ai_total_tokens', '$ai_reasoning_tokens']);
+
+/**
+ * Whether the caller supplied their own token counts, which override the ones the SDK
+ * derived from the provider response.
+ */
+function hasTokenOverrides(posthogProperties) {
+  return !!posthogProperties && Object.keys(posthogProperties).some(key => TOKEN_PROPERTY_KEYS.has(key));
+}
 function getTokensSource(posthogProperties) {
-  if (posthogProperties && Object.keys(posthogProperties).some(key => TOKEN_PROPERTY_KEYS.has(key))) {
-    return 'passthrough';
-  }
-  return 'sdk';
+  return hasTokenOverrides(posthogProperties) ? 'passthrough' : 'sdk';
 }
 
 // limit large outputs by truncating to 200kb (approx 200k bytes)
@@ -406,7 +411,7 @@ function sanitizeValues(obj) {
   return jsonSafe;
 }
 
-var version = "8.7.0";
+var version = "8.7.1";
 
 const DEFAULT_MAX_DEPTH = 3;
 const MAX_STACK_LINES = 20;
@@ -573,6 +578,10 @@ const captureAiGeneration = async (client, options) => {
         $ai_total_cost_usd: inputCostUSD + outputCostUSD
       };
     }
+
+    // The caller's own token counts override the SDK-derived ones further down, via the
+    // `options.properties` spread.
+    const tokensOverridden = hasTokenOverrides(options.properties);
     const additionalTokenValues = {
       ...(usage.reasoningTokens ? {
         $ai_reasoning_tokens: usage.reasoningTokens
@@ -583,6 +592,18 @@ const captureAiGeneration = async (client, options) => {
       ...(usage.cacheCreationInputTokens ? {
         $ai_cache_creation_input_tokens: usage.cacheCreationInputTokens
       } : {}),
+      // Checked against undefined rather than truthiness, because false is the meaningful
+      // value here and a truthiness guard would drop it.
+      //
+      // Dropped entirely when the caller overrides the token counts: the flag describes how
+      // the SDK-derived counts relate to each other, so against passthrough counts it can be
+      // wrong in the expensive direction. Declaring inclusive over counts that are actually
+      // exclusive makes ingestion subtract the cache pool that was never in the input. A
+      // caller who knows their own accounting model can still pass
+      // `$ai_cache_reporting_exclusive` themselves, and that value wins.
+      ...(usage.cacheReportingExclusive !== undefined && !tokensOverridden ? {
+        $ai_cache_reporting_exclusive: usage.cacheReportingExclusive
+      } : {}),
       ...(usage.webSearchCount ? {
         $ai_web_search_count: usage.webSearchCount
       } : {}),