@posthog/ai 7.16.15 → 7.17.1

package/dist/openai/index.mjs

@@ -2,8 +2,6 @@ import { OpenAI } from 'openai';
 import { v4 } from 'uuid';
 import { uuidv7 } from '@posthog/core';
 
-var version = "7.16.15";
-
 // Type guards for safer type checking
 
 const isString = value => {
@@ -164,6 +162,14 @@ function getTokensSource(posthogProperties) {
   }
   return 'sdk';
 }
+const STRING_FORMAT = 'utf8';
+
+// Reused across calls to avoid per-invocation allocation; truncate() runs
+// hundreds of times for prompts with many parts.
+new TextEncoder();
+new TextDecoder(STRING_FORMAT, {
+  fatal: false
+});
 
 /**
  * Safely converts content to a string, preserving structure for objects/arrays.
@@ -476,73 +482,114 @@ function addDefaults(params) {
     traceId: params.traceId ?? v4()
   };
 }
-const sendEventWithErrorToPosthog = async ({
-  client,
-  traceId,
-  error,
-  ...args
-}) => {
-  const httpStatus = error && typeof error === 'object' && 'status' in error ? error.status ?? 500 : 500;
-  const properties = {
-    client,
-    traceId,
-    httpStatus,
-    error: JSON.stringify(error),
-    ...args
-  };
-  const enrichedError = error;
-  if (client.options?.enableExceptionAutocapture) {
-    // assign a uuid that can be used to link the trace and exception events
-    const exceptionId = uuidv7();
-    client.captureException(error, undefined, {
-      $ai_trace_id: traceId
-    }, exceptionId);
-    enrichedError.__posthog_previously_captured_error = true;
-    properties.exceptionId = exceptionId;
+function formatOpenAIResponsesInput(input, instructions) {
+  const messages = [];
+  if (instructions) {
+    messages.push({
+      role: 'system',
+      content: instructions
+    });
   }
-  await sendEventToPosthog(properties);
-  return enrichedError;
-};
-const sendEventToPosthog = async ({
-  client,
-  eventType = AIEvent.Generation,
-  distinctId,
-  traceId,
-  model,
-  provider,
-  input,
-  output,
-  latency,
-  timeToFirstToken,
-  baseURL,
-  params,
-  httpStatus = 200,
-  usage = {},
-  error,
-  exceptionId,
-  stopReason,
-  tools,
-  captureImmediate = false
-}) => {
+  if (Array.isArray(input)) {
+    for (const item of input) {
+      if (typeof item === 'string') {
+        messages.push({
+          role: 'user',
+          content: item
+        });
+      } else if (item && typeof item === 'object') {
+        const obj = item;
+        const role = isString(obj.role) ? obj.role : 'user';
+
+        // Handle content properly - preserve structure for objects/arrays
+        const content = obj.content ?? obj.text ?? item;
+        messages.push({
+          role,
+          content: toContentString(content)
+        });
+      } else {
+        messages.push({
+          role: 'user',
+          content: toContentString(item)
+        });
+      }
+    }
+  } else if (typeof input === 'string') {
+    messages.push({
+      role: 'user',
+      content: input
+    });
+  } else if (input) {
+    messages.push({
+      role: 'user',
+      content: toContentString(input)
+    });
+  }
+  return messages;
+}
+
+var version = "7.17.1";
+
+/**
+ * Options for `captureAiGeneration`. Mirrors the `$ai_generation` event shape
+ * directly so that any caller — first-party SDK wrappers and external code
+ * alike — produces an identical event.
+ */
+
+/**
+ * Capture an `$ai_generation` (or `$ai_embedding`) event to PostHog.
+ *
+ * This is the canonical primitive that every `@posthog/ai` wrapper
+ * (`withTracing`, `OpenAI`, `Anthropic`, `GoogleGenAI`, …) funnels through, so
+ * external code can use it directly to instrument LLM calls made through
+ * arbitrary clients (Cloudflare Workers AI, custom HTTP, etc.) and get the
+ * same events the SDK wrappers produce.
+ *
+ * When `error` is set, the event is captured as an error. If the error is an
+ * object, it is mutated in place to set `__posthog_previously_captured_error`
+ * so callers can re-throw the original error reference safely.
+ */
+const captureAiGeneration = async (client, options) => {
   if (!client.capture) {
-    return Promise.resolve();
+    return;
   }
-  // sanitize input and output for UTF-8 validity
-  const safeInput = sanitizeValues(input);
-  const safeOutput = sanitizeValues(output);
-  const safeError = sanitizeValues(error);
+  const traceId = options.traceId ?? v4();
+  const eventType = options.eventType ?? AIEvent.Generation;
+  const privacyMode = options.privacyMode ?? false;
+  const usage = options.usage ?? {};
+  const safeInput = sanitizeValues(options.input);
+  const safeOutput = sanitizeValues(options.output);
+  let httpStatus = options.httpStatus;
   let errorData = {};
-  if (error) {
+  if (options.error) {
+    if (httpStatus === undefined) {
+      if (typeof options.error === 'object' && 'status' in options.error && typeof options.error.status === 'number') {
+        httpStatus = options.error.status;
+      } else {
+        httpStatus = 500;
+      }
+    }
+    let exceptionId;
+    if (client.options?.enableExceptionAutocapture) {
+      exceptionId = uuidv7();
+      client.captureException(options.error, undefined, {
+        $ai_trace_id: traceId
+      }, exceptionId);
+      if (typeof options.error === 'object') {
+        options.error.__posthog_previously_captured_error = true;
+      }
+    }
     errorData = {
       $ai_is_error: true,
-      $ai_error: safeError,
+      $ai_error: sanitizeValues(JSON.stringify(options.error)),
       $exception_event_id: exceptionId
     };
   }
+  httpStatus = httpStatus ?? 200;
   let costOverrideData = {};
-  if (params.posthogCostOverride) {
-    const inputCostUSD = (params.posthogCostOverride.inputCost ?? 0) * (usage.inputTokens ?? 0);
-    const outputCostUSD = (params.posthogCostOverride.outputCost ?? 0) * (usage.outputTokens ?? 0);
+  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,
@@ -569,96 +616,49 @@ const sendEventToPosthog = async ({
   const properties = {
     $ai_lib: 'posthog-ai',
     $ai_lib_version: version,
-    $ai_provider: params.posthogProviderOverride ?? provider,
-    $ai_model: params.posthogModelOverride ?? model,
-    $ai_model_parameters: getModelParams(params),
-    $ai_input: withPrivacyMode(client, params.posthogPrivacyMode ?? false, safeInput),
-    $ai_output_choices: withPrivacyMode(client, params.posthogPrivacyMode ?? false, safeOutput),
+    $ai_provider: options.providerOverride ?? options.provider,
+    $ai_model: options.modelOverride ?? options.model,
+    $ai_model_parameters: options.modelParameters ?? {},
+    $ai_input: withPrivacyMode(client, privacyMode, safeInput),
+    $ai_output_choices: withPrivacyMode(client, privacyMode, safeOutput),
     $ai_http_status: httpStatus,
     $ai_input_tokens: usage.inputTokens ?? 0,
     ...(usage.outputTokens !== undefined ? {
       $ai_output_tokens: usage.outputTokens
     } : {}),
     ...additionalTokenValues,
-    $ai_latency: latency,
-    ...(timeToFirstToken !== undefined ? {
-      $ai_time_to_first_token: timeToFirstToken
+    $ai_latency: options.latency ?? 0,
+    ...(options.timeToFirstToken !== undefined ? {
+      $ai_time_to_first_token: options.timeToFirstToken
     } : {}),
     $ai_trace_id: traceId,
-    $ai_base_url: baseURL,
-    ...params.posthogProperties,
-    $ai_tokens_source: getTokensSource(params.posthogProperties),
-    ...(distinctId ? {} : {
+    $ai_base_url: options.baseURL ?? '',
+    ...options.properties,
+    $ai_tokens_source: getTokensSource(options.properties),
+    ...(options.distinctId ? {} : {
       $process_person_profile: false
     }),
-    ...(stopReason ? {
-      $ai_stop_reason: stopReason
+    ...(options.stopReason ? {
+      $ai_stop_reason: options.stopReason
     } : {}),
-    ...(tools ? {
-      $ai_tools: tools
+    ...(options.tools ? {
+      $ai_tools: options.tools
     } : {}),
     ...errorData,
     ...costOverrideData
   };
   const event = {
-    distinctId: distinctId ?? traceId,
+    distinctId: options.distinctId ?? traceId,
     event: eventType,
     properties,
-    groups: params.posthogGroups
+    groups: options.groups
   };
-  if (captureImmediate) {
-    // await capture promise to send single event in serverless environments
+  if (options.captureImmediate) {
     await client.captureImmediate(event);
   } else {
     client.capture(event);
   }
-  return Promise.resolve();
 };
-function formatOpenAIResponsesInput(input, instructions) {
-  const messages = [];
-  if (instructions) {
-    messages.push({
-      role: 'system',
-      content: instructions
-    });
-  }
-  if (Array.isArray(input)) {
-    for (const item of input) {
-      if (typeof item === 'string') {
-        messages.push({
-          role: 'user',
-          content: item
-        });
-      } else if (item && typeof item === 'object') {
-        const obj = item;
-        const role = isString(obj.role) ? obj.role : 'user';
-
-        // Handle content properly - preserve structure for objects/arrays
-        const content = obj.content ?? obj.text ?? item;
-        messages.push({
-          role,
-          content: toContentString(content)
-        });
-      } else {
-        messages.push({
-          role: 'user',
-          content: toContentString(item)
-        });
-      }
-    }
-  } else if (typeof input === 'string') {
-    messages.push({
-      role: 'user',
-      content: input
-    });
-  } else if (input) {
-    messages.push({
-      role: 'user',
-      content: toContentString(input)
-    });
-  }
-  return messages;
-}
 
 /**
  * Checks if a ResponseStreamEvent chunk represents the first token/content from the model.
@@ -841,8 +841,7 @@ class WrappedCompletions extends Completions {
               const latency = (Date.now() - startTime) / 1000;
               const timeToFirstToken = firstTokenTime !== undefined ? (firstTokenTime - startTime) / 1000 : undefined;
               const availableTools = extractAvailableToolCalls('openai', openAIParams);
-              await sendEventToPosthog({
-                client: this.phClient,
+              await captureAiGeneration(this.phClient, {
                 ...posthogParams,
                 model: openAIParams.model ?? modelFromResponse,
                 provider: 'openai',
@@ -851,7 +850,7 @@ class WrappedCompletions extends Completions {
                 latency,
                 timeToFirstToken,
                 baseURL: this.baseURL,
-                params: body,
+                modelParameters: getModelParams(body),
                 httpStatus: 200,
                 usage: {
                   inputTokens: usage.inputTokens,
@@ -865,8 +864,7 @@ class WrappedCompletions extends Completions {
                 tools: availableTools
               });
             } catch (error) {
-              const enrichedError = await sendEventWithErrorToPosthog({
-                client: this.phClient,
+              await captureAiGeneration(this.phClient, {
                 ...posthogParams,
                 model: openAIParams.model,
                 provider: 'openai',
@@ -874,14 +872,14 @@ class WrappedCompletions extends Completions {
                 output: [],
                 latency: 0,
                 baseURL: this.baseURL,
-                params: body,
+                modelParameters: getModelParams(body),
                 usage: {
                   inputTokens: 0,
                   outputTokens: 0
                 },
                 error
               });
-              throw enrichedError;
+              throw error;
             }
           })();
 
@@ -896,8 +894,7 @@ class WrappedCompletions extends Completions {
           const latency = (Date.now() - startTime) / 1000;
           const availableTools = extractAvailableToolCalls('openai', openAIParams);
           const formattedOutput = formatResponseOpenAI(result);
-          await sendEventToPosthog({
-            client: this.phClient,
+          await captureAiGeneration(this.phClient, {
             ...posthogParams,
             model: openAIParams.model ?? result.model,
             provider: 'openai',
@@ -905,7 +902,7 @@ class WrappedCompletions extends Completions {
             output: formattedOutput,
             latency,
             baseURL: this.baseURL,
-            params: body,
+            modelParameters: getModelParams(body),
             httpStatus: 200,
             usage: {
               inputTokens: result.usage?.prompt_tokens ?? 0,
@@ -922,8 +919,7 @@ class WrappedCompletions extends Completions {
         return result;
       }, async error => {
         const httpStatus = error && typeof error === 'object' && 'status' in error ? error.status ?? 500 : 500;
-        await sendEventToPosthog({
-          client: this.phClient,
+        await captureAiGeneration(this.phClient, {
           ...posthogParams,
           model: openAIParams.model,
           provider: 'openai',
@@ -931,13 +927,13 @@ class WrappedCompletions extends Completions {
           output: [],
           latency: 0,
           baseURL: this.baseURL,
-          params: body,
+          modelParameters: getModelParams(body),
           httpStatus,
           usage: {
             inputTokens: 0,
             outputTokens: 0
           },
-          error: JSON.stringify(error)
+          error
         });
         throw error;
       });
@@ -1017,8 +1013,7 @@ class WrappedResponses extends Responses {
               const latency = (Date.now() - startTime) / 1000;
               const timeToFirstToken = firstTokenTime !== undefined ? (firstTokenTime - startTime) / 1000 : undefined;
               const availableTools = extractAvailableToolCalls('openai', openAIParams);
-              await sendEventToPosthog({
-                client: this.phClient,
+              await captureAiGeneration(this.phClient, {
                 ...posthogParams,
                 model: openAIParams.model ?? modelFromResponse,
                 provider: 'openai',
@@ -1027,7 +1022,7 @@ class WrappedResponses extends Responses {
                 latency,
                 timeToFirstToken,
                 baseURL: this.baseURL,
-                params: body,
+                modelParameters: getModelParams(body),
                 httpStatus: 200,
                 usage: {
                   inputTokens: usage.inputTokens,
@@ -1041,8 +1036,7 @@ class WrappedResponses extends Responses {
                 tools: availableTools
               });
             } catch (error) {
-              const enrichedError = await sendEventWithErrorToPosthog({
-                client: this.phClient,
+              await captureAiGeneration(this.phClient, {
                 ...posthogParams,
                 model: openAIParams.model,
                 provider: 'openai',
@@ -1050,14 +1044,14 @@ class WrappedResponses extends Responses {
                 output: [],
                 latency: 0,
                 baseURL: this.baseURL,
-                params: body,
+                modelParameters: getModelParams(body),
                 usage: {
                   inputTokens: 0,
                   outputTokens: 0
                 },
-                error: error
+                error
               });
-              throw enrichedError;
+              throw error;
             }
           })();
           return stream2;
@@ -1072,8 +1066,7 @@ class WrappedResponses extends Responses {
           const formattedOutput = formatResponseOpenAI({
             output: result.output
           });
-          await sendEventToPosthog({
-            client: this.phClient,
+          await captureAiGeneration(this.phClient, {
             ...posthogParams,
             model: openAIParams.model ?? result.model,
             provider: 'openai',
@@ -1081,7 +1074,7 @@ class WrappedResponses extends Responses {
             output: formattedOutput,
             latency,
             baseURL: this.baseURL,
-            params: body,
+            modelParameters: getModelParams(body),
             httpStatus: 200,
             usage: {
               inputTokens: result.usage?.input_tokens ?? 0,
@@ -1098,8 +1091,7 @@ class WrappedResponses extends Responses {
         return result;
       }, async error => {
         const httpStatus = error && typeof error === 'object' && 'status' in error ? error.status ?? 500 : 500;
-        await sendEventToPosthog({
-          client: this.phClient,
+        await captureAiGeneration(this.phClient, {
           ...posthogParams,
           model: openAIParams.model,
           provider: 'openai',
@@ -1107,13 +1099,13 @@ class WrappedResponses extends Responses {
           output: [],
           latency: 0,
           baseURL: this.baseURL,
-          params: body,
+          modelParameters: getModelParams(body),
           httpStatus,
           usage: {
             inputTokens: 0,
             outputTokens: 0
           },
-          error: JSON.stringify(error)
+          error
         });
         throw error;
       });
@@ -1134,8 +1126,7 @@ class WrappedResponses extends Responses {
       const parentPromise = super.parse(openAIParams, options);
       const wrappedPromise = parentPromise.then(async result => {
         const latency = (Date.now() - startTime) / 1000;
-        await sendEventToPosthog({
-          client: this.phClient,
+        await captureAiGeneration(this.phClient, {
           ...posthogParams,
           model: openAIParams.model ?? result.model,
           provider: 'openai',
@@ -1143,7 +1134,7 @@ class WrappedResponses extends Responses {
           output: result.output,
           latency,
           baseURL: this.baseURL,
-          params: body,
+          modelParameters: getModelParams(body),
           httpStatus: 200,
           usage: {
             inputTokens: result.usage?.input_tokens ?? 0,
@@ -1156,8 +1147,7 @@ class WrappedResponses extends Responses {
         });
         return result;
       }, async error => {
-        const enrichedError = await sendEventWithErrorToPosthog({
-          client: this.phClient,
+        await captureAiGeneration(this.phClient, {
           ...posthogParams,
           model: openAIParams.model,
           provider: 'openai',
@@ -1165,14 +1155,14 @@ class WrappedResponses extends Responses {
           output: [],
           latency: 0,
           baseURL: this.baseURL,
-          params: body,
+          modelParameters: getModelParams(body),
           usage: {
             inputTokens: 0,
             outputTokens: 0
           },
-          error: JSON.stringify(error)
+          error
         });
-        throw enrichedError;
+        throw error;
       });
       return wrappedPromise;
     } finally {
@@ -1196,8 +1186,7 @@ class WrappedEmbeddings extends Embeddings {
     const parentPromise = super.create(openAIParams, options);
     const wrappedPromise = parentPromise.then(async result => {
       const latency = (Date.now() - startTime) / 1000;
-      await sendEventToPosthog({
-        client: this.phClient,
+      await captureAiGeneration(this.phClient, {
         ...posthogParams,
         eventType: AIEvent.Embedding,
         model: openAIParams.model,
@@ -1207,7 +1196,7 @@ class WrappedEmbeddings extends Embeddings {
         // Embeddings don't have output content
         latency,
         baseURL: this.baseURL,
-        params: body,
+        modelParameters: getModelParams(body),
         httpStatus: 200,
         usage: {
           inputTokens: result.usage?.prompt_tokens ?? 0,
@@ -1217,8 +1206,7 @@ class WrappedEmbeddings extends Embeddings {
       return result;
     }, async error => {
       const httpStatus = error && typeof error === 'object' && 'status' in error ? error.status ?? 500 : 500;
-      await sendEventToPosthog({
-        client: this.phClient,
+      await captureAiGeneration(this.phClient, {
         eventType: AIEvent.Embedding,
         ...posthogParams,
         model: openAIParams.model,
@@ -1228,12 +1216,12 @@ class WrappedEmbeddings extends Embeddings {
         // Embeddings don't have output content
         latency: 0,
         baseURL: this.baseURL,
-        params: body,
+        modelParameters: getModelParams(body),
         httpStatus,
         usage: {
           inputTokens: 0
         },
-        error: JSON.stringify(error)
+        error
       });
       throw error;
     });
@@ -1307,8 +1295,7 @@ class WrappedTranscriptions extends Transcriptions {
               const latency = (Date.now() - startTime) / 1000;
               const timeToFirstToken = firstTokenTime !== undefined ? (firstTokenTime - startTime) / 1000 : undefined;
               const availableTools = extractAvailableToolCalls('openai', openAIParams);
-              await sendEventToPosthog({
-                client: this.phClient,
+              await captureAiGeneration(this.phClient, {
                 ...posthogParams,
                 model: openAIParams.model,
                 provider: 'openai',
@@ -1317,14 +1304,13 @@ class WrappedTranscriptions extends Transcriptions {
                 latency,
                 timeToFirstToken,
                 baseURL: this.baseURL,
-                params: body,
+                modelParameters: getModelParams(body),
                 httpStatus: 200,
                 usage,
                 tools: availableTools
               });
             } catch (error) {
-              const enrichedError = await sendEventWithErrorToPosthog({
-                client: this.phClient,
+              await captureAiGeneration(this.phClient, {
                 ...posthogParams,
                 model: openAIParams.model,
                 provider: 'openai',
@@ -1332,14 +1318,14 @@ class WrappedTranscriptions extends Transcriptions {
                 output: [],
                 latency: 0,
                 baseURL: this.baseURL,
-                params: body,
+                modelParameters: getModelParams(body),
                 usage: {
                   inputTokens: 0,
                   outputTokens: 0
                 },
-                error: error
+                error
               });
-              throw enrichedError;
+              throw error;
             }
           })();
           return stream2;
@@ -1350,8 +1336,7 @@ class WrappedTranscriptions extends Transcriptions {
       const wrappedPromise = parentPromise.then(async result => {
         if ('text' in result) {
           const latency = (Date.now() - startTime) / 1000;
-          await sendEventToPosthog({
-            client: this.phClient,
+          await captureAiGeneration(this.phClient, {
             ...posthogParams,
             model: openAIParams.model,
             provider: 'openai',
@@ -1359,7 +1344,7 @@ class WrappedTranscriptions extends Transcriptions {
             output: result.text,
             latency,
             baseURL: this.baseURL,
-            params: body,
+            modelParameters: getModelParams(body),
             httpStatus: 200,
             usage: {
               inputTokens: result.usage?.type === 'tokens' ? result.usage.input_tokens ?? 0 : 0,
@@ -1370,8 +1355,7 @@ class WrappedTranscriptions extends Transcriptions {
           return result;
         }
       }, async error => {
-        const enrichedError = await sendEventWithErrorToPosthog({
-          client: this.phClient,
+        await captureAiGeneration(this.phClient, {
           ...posthogParams,
           model: openAIParams.model,
           provider: 'openai',
@@ -1379,14 +1363,14 @@ class WrappedTranscriptions extends Transcriptions {
           output: [],
           latency: 0,
           baseURL: this.baseURL,
-          params: body,
+          modelParameters: getModelParams(body),
           usage: {
             inputTokens: 0,
             outputTokens: 0
           },
-          error: error
+          error
         });
-        throw enrichedError;
+        throw error;
       });
       return wrappedPromise;
     }