@posthog/ai 8.9.1 → 8.9.2

package/dist/anthropic/index.cjs

@@ -338,7 +338,7 @@ function addDefaults(params) {
   };
 }
 
-var version = "8.9.1";
+var version = "8.9.2";
 
 const DEFAULT_MAX_DEPTH = 3;
 const MAX_STACK_LINES = 20;
@@ -495,15 +495,22 @@ const captureAiGeneration = async (client, options) => {
       };
     }
     httpStatus = httpStatus ?? 200;
-    let costOverrideData = {};
+
+    // A configured price applies only to a count the provider reported, so a call with no
+    // reported usage sends no cost instead of asserting $0. $ai_total_cost_usd sums the sides
+    // that were priced, which makes it the cost of the known side alone when the other side
+    // went unreported: a lower bound on the true total, not an assertion of it.
+    const costOverrideData = {};
     if (options.costOverride) {
-      const inputCostUSD = (options.costOverride.inputCost ?? 0) * (usage.inputTokens ?? 0);
-      const outputCostUSD = (options.costOverride.outputCost ?? 0) * (usage.outputTokens ?? 0);
-      costOverrideData = {
-        $ai_input_cost_usd: inputCostUSD,
-        $ai_output_cost_usd: outputCostUSD,
-        $ai_total_cost_usd: inputCostUSD + outputCostUSD
-      };
+      if (usage.inputTokens !== undefined) {
+        costOverrideData.$ai_input_cost_usd = (options.costOverride.inputCost ?? 0) * usage.inputTokens;
+      }
+      if (usage.outputTokens !== undefined) {
+        costOverrideData.$ai_output_cost_usd = (options.costOverride.outputCost ?? 0) * usage.outputTokens;
+      }
+      if (Object.keys(costOverrideData).length > 0) {
+        costOverrideData.$ai_total_cost_usd = (costOverrideData.$ai_input_cost_usd ?? 0) + (costOverrideData.$ai_output_cost_usd ?? 0);
+      }
     }
 
     // The caller's own token counts override the SDK-derived ones further down, via the
@@ -547,7 +554,9 @@ const captureAiGeneration = async (client, options) => {
       $ai_input: safeInput,
       $ai_output_choices: safeOutput,
       $ai_http_status: httpStatus,
-      $ai_input_tokens: usage.inputTokens ?? 0,
+      ...(usage.inputTokens !== undefined ? {
+        $ai_input_tokens: usage.inputTokens
+      } : {}),
       ...(usage.outputTokens !== undefined ? {
         $ai_output_tokens: usage.outputTokens
       } : {}),
@@ -927,10 +936,6 @@ class WrappedMessages extends AnthropicOriginal__default.default.Messages {
         let firstTokenTime;
         let stopReason;
         const usage = {
-          inputTokens: 0,
-          outputTokens: 0,
-          cacheCreationInputTokens: 0,
-          cacheReadInputTokens: 0,
           webSearchCount: 0
         };
         let rawUsage = {};
@@ -1074,19 +1079,21 @@ class WrappedMessages extends AnthropicOriginal__default.default.Messages {
                 tools: availableTools
               });
             } catch (error) {
+              // The final usage delta may never arrive; whatever raw usage did
+              // still belongs on the event.
+              if (Object.keys(rawUsage).length > 0) {
+                usage.rawUsage = rawUsage;
+              }
               await captureAiGeneration(this.phClient, {
                 ...posthogParams,
                 model: anthropicParams.model,
                 provider: 'anthropic',
                 input: sanitizeAnthropic(mergeSystemPrompt(anthropicParams), this.phClient),
                 output: [],
-                latency: 0,
+                latency: (Date.now() - startTime) / 1000,
                 baseURL: this.baseURL,
                 modelParameters: getModelParams(body),
-                usage: {
-                  inputTokens: 0,
-                  outputTokens: 0
-                },
+                usage,
                 error: error
               });
               throw error;
@@ -1141,14 +1148,11 @@ class WrappedMessages extends AnthropicOriginal__default.default.Messages {
           provider: 'anthropic',
           input: sanitizeAnthropic(mergeSystemPrompt(anthropicParams), this.phClient),
           output: [],
-          latency: 0,
+          latency: (Date.now() - startTime) / 1000,
           baseURL: this.baseURL,
           modelParameters: getModelParams(body),
           httpStatus: error?.status ? error.status : 500,
-          usage: {
-            inputTokens: 0,
-            outputTokens: 0
-          },
+          usage: {},
           error: error
         });
         throw error;