@posthog/ai 7.16.15 → 7.17.1

package/dist/index.mjs

@@ -5,8 +5,6 @@ import { uuidv7 } from '@posthog/core';
 import AnthropicOriginal from '@anthropic-ai/sdk';
 import { GoogleGenAI } from '@google/genai';
 
-var version = "7.16.15";
-
 // Type guards for safer type checking
 const isString = value => {
   return typeof value === 'string';
@@ -244,6 +242,13 @@ function getTokensSource(posthogProperties) {
 // limit large outputs by truncating to 200kb (approx 200k bytes)
 const MAX_OUTPUT_SIZE = 200000;
 const STRING_FORMAT = 'utf8';
+// Reused across calls to avoid per-invocation allocation; truncate() runs
+// hundreds of times for prompts with many parts.
+const sharedTextEncoder = new TextEncoder();
+const sharedTextDecoder = new TextDecoder(STRING_FORMAT, {
+  fatal: false
+});
+const utf8ByteLength = str => sharedTextEncoder.encode(str).byteLength;
 /**
  * Safely converts content to a string, preserving structure for objects/arrays.
  * - If content is already a string, returns it as-is
@@ -514,19 +519,15 @@ const truncate = input => {
     return '';
   }
   // Check if we need to truncate and ensure STRING_FORMAT is respected
-  const encoder = new TextEncoder();
-  const buffer = encoder.encode(str);
+  const buffer = sharedTextEncoder.encode(str);
   if (buffer.length <= MAX_OUTPUT_SIZE) {
     // Ensure STRING_FORMAT is respected
-    return new TextDecoder(STRING_FORMAT).decode(buffer);
+    return sharedTextDecoder.decode(buffer);
   }
-  // Truncate the buffer and ensure a valid string is returned
+  // Truncate the buffer and ensure a valid string is returned.
+  // fatal: false means we get U+FFFD at the end if truncation broke the encoding.
   const truncatedBuffer = buffer.slice(0, MAX_OUTPUT_SIZE);
-  // fatal: false means we get U+FFFD at the end if truncation broke the encoding
-  const decoder = new TextDecoder(STRING_FORMAT, {
-    fatal: false
-  });
-  let truncatedStr = decoder.decode(truncatedBuffer);
+  let truncatedStr = sharedTextDecoder.decode(truncatedBuffer);
   if (truncatedStr.endsWith('\uFFFD')) {
     truncatedStr = truncatedStr.slice(0, -1);
   }
@@ -710,73 +711,107 @@ 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;
-  }
-  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 (!client.capture) {
-    return Promise.resolve();
+function formatOpenAIResponsesInput(input, instructions) {
+  const messages = [];
+  if (instructions) {
+    messages.push({
+      role: 'system',
+      content: instructions
+    });
   }
-  // sanitize input and output for UTF-8 validity
-  const safeInput = sanitizeValues(input);
-  const safeOutput = sanitizeValues(output);
-  const safeError = sanitizeValues(error);
+  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";
+
+/**
+ * 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;
+  }
+  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,
@@ -803,95 +838,49 @@ 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.
@@ -1060,8 +1049,7 @@ let WrappedCompletions$1 = 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',
@@ -1070,7 +1058,7 @@ let WrappedCompletions$1 = class WrappedCompletions extends Completions {
                 latency,
                 timeToFirstToken,
                 baseURL: this.baseURL,
-                params: body,
+                modelParameters: getModelParams(body),
                 httpStatus: 200,
                 usage: {
                   inputTokens: usage.inputTokens,
@@ -1084,8 +1072,7 @@ let WrappedCompletions$1 = 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',
@@ -1093,14 +1080,14 @@ let WrappedCompletions$1 = 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;
             }
           })();
           // Return the other stream to the user
@@ -1114,8 +1101,7 @@ let WrappedCompletions$1 = 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',
@@ -1123,7 +1109,7 @@ let WrappedCompletions$1 = class WrappedCompletions extends Completions {
             output: formattedOutput,
             latency,
             baseURL: this.baseURL,
-            params: body,
+            modelParameters: getModelParams(body),
             httpStatus: 200,
             usage: {
               inputTokens: result.usage?.prompt_tokens ?? 0,
@@ -1140,8 +1126,7 @@ let WrappedCompletions$1 = 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',
@@ -1149,13 +1134,13 @@ let WrappedCompletions$1 = 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;
       });
@@ -1228,8 +1213,7 @@ let WrappedResponses$1 = 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',
@@ -1238,7 +1222,7 @@ let WrappedResponses$1 = class WrappedResponses extends Responses {
                 latency,
                 timeToFirstToken,
                 baseURL: this.baseURL,
-                params: body,
+                modelParameters: getModelParams(body),
                 httpStatus: 200,
                 usage: {
                   inputTokens: usage.inputTokens,
@@ -1252,8 +1236,7 @@ let WrappedResponses$1 = 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',
@@ -1261,14 +1244,14 @@ let WrappedResponses$1 = 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;
@@ -1283,8 +1266,7 @@ let WrappedResponses$1 = 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',
@@ -1292,7 +1274,7 @@ let WrappedResponses$1 = class WrappedResponses extends Responses {
             output: formattedOutput,
             latency,
             baseURL: this.baseURL,
-            params: body,
+            modelParameters: getModelParams(body),
             httpStatus: 200,
             usage: {
               inputTokens: result.usage?.input_tokens ?? 0,
@@ -1309,8 +1291,7 @@ let WrappedResponses$1 = 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',
@@ -1318,13 +1299,13 @@ let WrappedResponses$1 = 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;
       });
@@ -1345,8 +1326,7 @@ let WrappedResponses$1 = 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',
@@ -1354,7 +1334,7 @@ let WrappedResponses$1 = 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,
@@ -1367,8 +1347,7 @@ let WrappedResponses$1 = 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',
@@ -1376,14 +1355,14 @@ let WrappedResponses$1 = 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 {
@@ -1407,8 +1386,7 @@ let WrappedEmbeddings$1 = 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,
@@ -1418,7 +1396,7 @@ let WrappedEmbeddings$1 = 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,
@@ -1428,8 +1406,7 @@ let WrappedEmbeddings$1 = 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,
@@ -1439,12 +1416,12 @@ let WrappedEmbeddings$1 = 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;
     });
@@ -1503,8 +1480,7 @@ 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',
@@ -1513,14 +1489,13 @@ 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',
@@ -1528,14 +1503,14 @@ 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;
@@ -1546,8 +1521,7 @@ 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',
@@ -1555,7 +1529,7 @@ 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,
@@ -1566,8 +1540,7 @@ 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',
@@ -1575,14 +1548,14 @@ 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;
     }
@@ -1727,8 +1700,7 @@ class WrappedCompletions extends AzureOpenAI.Chat.Completions {
               }];
               const latency = (Date.now() - startTime) / 1000;
               const timeToFirstToken = firstTokenTime !== undefined ? (firstTokenTime - startTime) / 1000 : undefined;
-              await sendEventToPosthog({
-                client: this.phClient,
+              await captureAiGeneration(this.phClient, {
                 ...posthogParams,
                 model: openAIParams.model ?? modelFromResponse,
                 provider: 'azure',
@@ -1737,13 +1709,12 @@ class WrappedCompletions extends AzureOpenAI.Chat.Completions {
                 latency,
                 timeToFirstToken,
                 baseURL: this.baseURL,
-                params: body,
+                modelParameters: getModelParams(body),
                 httpStatus: 200,
                 usage
               });
             } catch (error) {
-              const enrichedError = await sendEventWithErrorToPosthog({
-                client: this.phClient,
+              await captureAiGeneration(this.phClient, {
                 ...posthogParams,
                 model: openAIParams.model,
                 provider: 'azure',
@@ -1751,14 +1722,14 @@ class WrappedCompletions extends AzureOpenAI.Chat.Completions {
                 output: [],
                 latency: 0,
                 baseURL: this.baseURL,
-                params: body,
+                modelParameters: getModelParams(body),
                 usage: {
                   inputTokens: 0,
                   outputTokens: 0
                 },
                 error: error
               });
-              throw enrichedError;
+              throw error;
             }
           })();
           // Return the other stream to the user
@@ -1770,8 +1741,7 @@ class WrappedCompletions extends AzureOpenAI.Chat.Completions {
       const wrappedPromise = parentPromise.then(async result => {
         if ('choices' in result) {
           const latency = (Date.now() - startTime) / 1000;
-          await sendEventToPosthog({
-            client: this.phClient,
+          await captureAiGeneration(this.phClient, {
             ...posthogParams,
             model: openAIParams.model ?? result.model,
             provider: 'azure',
@@ -1779,7 +1749,7 @@ class WrappedCompletions extends AzureOpenAI.Chat.Completions {
             output: formatResponseOpenAI(result),
             latency,
             baseURL: this.baseURL,
-            params: body,
+            modelParameters: getModelParams(body),
             httpStatus: 200,
             usage: {
               inputTokens: result.usage?.prompt_tokens ?? 0,
@@ -1792,8 +1762,7 @@ class WrappedCompletions extends AzureOpenAI.Chat.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: 'azure',
@@ -1801,13 +1770,13 @@ class WrappedCompletions extends AzureOpenAI.Chat.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;
       });
@@ -1867,8 +1836,7 @@ class WrappedResponses extends AzureOpenAI.Responses {
               }
               const latency = (Date.now() - startTime) / 1000;
               const timeToFirstToken = firstTokenTime !== undefined ? (firstTokenTime - startTime) / 1000 : undefined;
-              await sendEventToPosthog({
-                client: this.phClient,
+              await captureAiGeneration(this.phClient, {
                 ...posthogParams,
                 model: openAIParams.model ?? modelFromResponse,
                 provider: 'azure',
@@ -1877,13 +1845,12 @@ class WrappedResponses extends AzureOpenAI.Responses {
                 latency,
                 timeToFirstToken,
                 baseURL: this.baseURL,
-                params: body,
+                modelParameters: getModelParams(body),
                 httpStatus: 200,
                 usage
               });
             } catch (error) {
-              const enrichedError = await sendEventWithErrorToPosthog({
-                client: this.phClient,
+              await captureAiGeneration(this.phClient, {
                 ...posthogParams,
                 model: openAIParams.model,
                 provider: 'azure',
@@ -1891,14 +1858,14 @@ class WrappedResponses extends AzureOpenAI.Responses {
                 output: [],
                 latency: 0,
                 baseURL: this.baseURL,
-                params: body,
+                modelParameters: getModelParams(body),
                 usage: {
                   inputTokens: 0,
                   outputTokens: 0
                 },
                 error: error
               });
-              throw enrichedError;
+              throw error;
             }
           })();
           return stream2;
@@ -1909,8 +1876,7 @@ class WrappedResponses extends AzureOpenAI.Responses {
       const wrappedPromise = parentPromise.then(async result => {
         if ('output' in result) {
           const latency = (Date.now() - startTime) / 1000;
-          await sendEventToPosthog({
-            client: this.phClient,
+          await captureAiGeneration(this.phClient, {
             ...posthogParams,
             model: openAIParams.model ?? result.model,
             provider: 'azure',
@@ -1918,7 +1884,7 @@ class WrappedResponses extends AzureOpenAI.Responses {
             output: result.output,
             latency,
             baseURL: this.baseURL,
-            params: body,
+            modelParameters: getModelParams(body),
             httpStatus: 200,
             usage: {
               inputTokens: result.usage?.input_tokens ?? 0,
@@ -1931,8 +1897,7 @@ class WrappedResponses extends AzureOpenAI.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: 'azure',
@@ -1940,13 +1905,13 @@ class WrappedResponses extends AzureOpenAI.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;
       });
@@ -1962,8 +1927,7 @@ class WrappedResponses extends AzureOpenAI.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: 'azure',
@@ -1971,7 +1935,7 @@ class WrappedResponses extends AzureOpenAI.Responses {
         output: result.output,
         latency,
         baseURL: this.baseURL,
-        params: body,
+        modelParameters: getModelParams(body),
         httpStatus: 200,
         usage: {
           inputTokens: result.usage?.input_tokens ?? 0,
@@ -1982,8 +1946,7 @@ class WrappedResponses extends AzureOpenAI.Responses {
       });
       return result;
     }, async error => {
-      await sendEventToPosthog({
-        client: this.phClient,
+      await captureAiGeneration(this.phClient, {
         ...posthogParams,
         model: openAIParams.model,
         provider: 'azure',
@@ -1991,13 +1954,13 @@ class WrappedResponses extends AzureOpenAI.Responses {
         output: [],
         latency: 0,
         baseURL: this.baseURL,
-        params: body,
+        modelParameters: getModelParams(body),
         httpStatus: error?.status ? error.status : 500,
         usage: {
           inputTokens: 0,
           outputTokens: 0
         },
-        error: JSON.stringify(error)
+        error
       });
       throw error;
     });
@@ -2019,8 +1982,7 @@ class WrappedEmbeddings extends AzureOpenAI.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, {
         eventType: AIEvent.Embedding,
         ...posthogParams,
         model: openAIParams.model,
@@ -2030,7 +1992,7 @@ class WrappedEmbeddings extends AzureOpenAI.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
@@ -2039,8 +2001,7 @@ class WrappedEmbeddings extends AzureOpenAI.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,
@@ -2049,12 +2010,12 @@ class WrappedEmbeddings extends AzureOpenAI.Embeddings {
         output: null,
         latency: 0,
         baseURL: this.baseURL,
-        params: body,
+        modelParameters: getModelParams(body),
         httpStatus,
         usage: {
           inputTokens: 0
         },
-        error: JSON.stringify(error)
+        error
       });
       throw error;
     });
@@ -2153,18 +2114,26 @@ const mapVercelPrompt = messages => {
     };
   });
   try {
-    // Trim the inputs array until its JSON size fits within MAX_OUTPUT_SIZE
-    const encoder = new TextEncoder();
-    let serialized = JSON.stringify(inputs);
+    // Trim the inputs array until its serialized JSON size fits within MAX_OUTPUT_SIZE.
+    // Pre-compute each message's byte size once so we can shift by accumulated budget
+    // in a single linear pass, instead of re-stringifying the whole array per iteration.
+    const messageSizes = inputs.map(m => utf8ByteLength(JSON.stringify(m)));
+    // Account for the surrounding `[` `]` plus a comma between each pair of elements.
+    let totalBytes = 2 + Math.max(0, messageSizes.length - 1);
+    for (const size of messageSizes) {
+      totalBytes += size;
+    }
     let removedCount = 0;
-    // We need to keep track of the initial size of the inputs array because we're going to be mutating it
-    const initialSize = inputs.length;
-    for (let i = 0; i < initialSize && encoder.encode(serialized).byteLength > MAX_OUTPUT_SIZE; i++) {
-      inputs.shift();
+    while (totalBytes > MAX_OUTPUT_SIZE && removedCount < messageSizes.length) {
+      totalBytes -= messageSizes[removedCount];
+      // Each removed message past the first also drops the comma that joined it.
+      if (removedCount < messageSizes.length - 1) {
+        totalBytes -= 1;
+      }
       removedCount++;
-      serialized = JSON.stringify(inputs);
     }
     if (removedCount > 0) {
+      inputs.splice(0, removedCount);
       // Add one placeholder to indicate how many were removed
       inputs.unshift({
         role: 'posthog',
@@ -2383,6 +2352,18 @@ const wrapVercelLanguageModel = (model, phClient, options) => {
       $ai_framework_version: model.specificationVersion === 'v3' ? '6' : '5'
     }
   };
+  // Shared `captureAiGeneration` options for every call site in this wrapper.
+  const baseOptions = {
+    distinctId: mergedOptions.posthogDistinctId,
+    traceId,
+    properties: mergedOptions.posthogProperties,
+    groups: mergedOptions.posthogGroups,
+    privacyMode: mergedOptions.posthogPrivacyMode,
+    modelOverride: mergedOptions.posthogModelOverride,
+    providerOverride: mergedOptions.posthogProviderOverride,
+    costOverride: mergedOptions.posthogCostOverride,
+    captureImmediate: mergedOptions.posthogCaptureImmediate
+  };
   // Create wrapped model using Object.create to preserve the prototype chain
   // This automatically inherits all properties (including getters) from the model
   const wrappedModel = Object.create(model, {
@@ -2435,46 +2416,40 @@ const wrapVercelLanguageModel = (model, phClient, options) => {
           // Extract finish reason - V2 returns a string, V3 returns an object with .unified
           const rawFinishReason = result.finishReason;
           const finishReasonStr = typeof rawFinishReason === 'string' ? rawFinishReason : rawFinishReason && typeof rawFinishReason === 'object' && 'unified' in rawFinishReason ? String(rawFinishReason.unified) : undefined;
-          await sendEventToPosthog({
-            client: phClient,
-            distinctId: mergedOptions.posthogDistinctId,
-            traceId: mergedOptions.posthogTraceId ?? v4(),
+          await captureAiGeneration(phClient, {
+            ...baseOptions,
             model: modelId,
             provider: provider,
             input: mergedOptions.posthogPrivacyMode ? '' : mapVercelPrompt(params.prompt),
             output: content,
             latency,
             baseURL,
-            params: mergedParams,
+            modelParameters: getModelParams(mergedParams),
             httpStatus: 200,
             usage,
             stopReason: finishReasonStr,
-            tools: availableTools,
-            captureImmediate: mergedOptions.posthogCaptureImmediate
+            tools: availableTools
           });
           return result;
         } catch (error) {
           const modelId = model.modelId;
-          const enrichedError = await sendEventWithErrorToPosthog({
-            client: phClient,
-            distinctId: mergedOptions.posthogDistinctId,
-            traceId: mergedOptions.posthogTraceId ?? v4(),
+          await captureAiGeneration(phClient, {
+            ...baseOptions,
             model: modelId,
             provider: model.provider,
             input: mergedOptions.posthogPrivacyMode ? '' : mapVercelPrompt(params.prompt),
             output: [],
             latency: 0,
             baseURL: '',
-            params: mergedParams,
+            modelParameters: getModelParams(mergedParams),
             usage: {
               inputTokens: 0,
               outputTokens: 0
             },
             error: error,
-            tools: availableTools,
-            captureImmediate: mergedOptions.posthogCaptureImmediate
+            tools: availableTools
           });
-          throw enrichedError;
+          throw error;
         }
       },
       writable: true,
@@ -2620,10 +2595,8 @@ const wrapVercelLanguageModel = (model, phClient, options) => {
                 }
               };
               adjustAnthropicV3CacheTokens(model, modelId, provider, finalUsage);
-              await sendEventToPosthog({
-                client: phClient,
-                distinctId: mergedOptions.posthogDistinctId,
-                traceId: mergedOptions.posthogTraceId ?? v4(),
+              await captureAiGeneration(phClient, {
+                ...baseOptions,
                 model: modelId,
                 provider: provider,
                 input: mergedOptions.posthogPrivacyMode ? '' : mapVercelPrompt(params.prompt),
@@ -2631,12 +2604,11 @@ const wrapVercelLanguageModel = (model, phClient, options) => {
                 latency,
                 timeToFirstToken,
                 baseURL,
-                params: mergedParams,
+                modelParameters: getModelParams(mergedParams),
                 httpStatus: 200,
                 usage: finalUsage,
                 stopReason,
-                tools: availableTools,
-                captureImmediate: mergedOptions.posthogCaptureImmediate
+                tools: availableTools
               });
             }
           });
@@ -2645,26 +2617,23 @@ const wrapVercelLanguageModel = (model, phClient, options) => {
             ...rest
           };
         } catch (error) {
-          const enrichedError = await sendEventWithErrorToPosthog({
-            client: phClient,
-            distinctId: mergedOptions.posthogDistinctId,
-            traceId: mergedOptions.posthogTraceId ?? v4(),
+          await captureAiGeneration(phClient, {
+            ...baseOptions,
             model: modelId,
             provider: provider,
             input: mergedOptions.posthogPrivacyMode ? '' : mapVercelPrompt(params.prompt),
             output: [],
             latency: 0,
             baseURL: '',
-            params: mergedParams,
+            modelParameters: getModelParams(mergedParams),
             usage: {
               inputTokens: 0,
               outputTokens: 0
             },
             error: error,
-            tools: availableTools,
-            captureImmediate: mergedOptions.posthogCaptureImmediate
+            tools: availableTools
           });
-          throw enrichedError;
+          throw error;
         }
       },
       writable: true,
@@ -2829,8 +2798,7 @@ class WrappedMessages extends AnthropicOriginal.Messages {
                   text: accumulatedContent
                 }]
               }];
-              await sendEventToPosthog({
-                client: this.phClient,
+              await captureAiGeneration(this.phClient, {
                 ...posthogParams,
                 model: anthropicParams.model,
                 provider: 'anthropic',
@@ -2839,15 +2807,14 @@ class WrappedMessages extends AnthropicOriginal.Messages {
                 latency,
                 timeToFirstToken,
                 baseURL: this.baseURL,
-                params: body,
+                modelParameters: getModelParams(body),
                 httpStatus: 200,
                 usage,
                 stopReason,
                 tools: availableTools
               });
             } catch (error) {
-              const enrichedError = await sendEventWithErrorToPosthog({
-                client: this.phClient,
+              await captureAiGeneration(this.phClient, {
                 ...posthogParams,
                 model: anthropicParams.model,
                 provider: 'anthropic',
@@ -2855,14 +2822,14 @@ class WrappedMessages extends AnthropicOriginal.Messages {
                 output: [],
                 latency: 0,
                 baseURL: this.baseURL,
-                params: body,
+                modelParameters: getModelParams(body),
                 usage: {
                   inputTokens: 0,
                   outputTokens: 0
                 },
                 error: error
               });
-              throw enrichedError;
+              throw error;
             }
           })();
           // Return the other stream to the user
@@ -2875,8 +2842,7 @@ class WrappedMessages extends AnthropicOriginal.Messages {
         if ('content' in result) {
           const latency = (Date.now() - startTime) / 1000;
           const availableTools = extractAvailableToolCalls('anthropic', anthropicParams);
-          await sendEventToPosthog({
-            client: this.phClient,
+          await captureAiGeneration(this.phClient, {
             ...posthogParams,
             model: anthropicParams.model,
             provider: 'anthropic',
@@ -2884,7 +2850,7 @@ class WrappedMessages extends AnthropicOriginal.Messages {
             output: formatResponseAnthropic(result),
             latency,
             baseURL: this.baseURL,
-            params: body,
+            modelParameters: getModelParams(body),
             httpStatus: 200,
             usage: {
               inputTokens: result.usage.input_tokens ?? 0,
@@ -2900,8 +2866,7 @@ class WrappedMessages extends AnthropicOriginal.Messages {
         }
         return result;
       }, async error => {
-        await sendEventToPosthog({
-          client: this.phClient,
+        await captureAiGeneration(this.phClient, {
           ...posthogParams,
           model: anthropicParams.model,
           provider: 'anthropic',
@@ -2909,13 +2874,13 @@ class WrappedMessages extends AnthropicOriginal.Messages {
           output: [],
           latency: 0,
           baseURL: this.baseURL,
-          params: body,
+          modelParameters: getModelParams(body),
           httpStatus: error?.status ? error.status : 500,
           usage: {
             inputTokens: 0,
             outputTokens: 0
           },
-          error: JSON.stringify(error)
+          error: error
         });
         throw error;
       });
@@ -2952,8 +2917,7 @@ class WrappedModels {
       const availableTools = extractAvailableToolCalls('gemini', geminiParams);
       const metadata = response.usageMetadata;
       const finishReason = response.candidates?.[0]?.finishReason;
-      await sendEventToPosthog({
-        client: this.phClient,
+      await captureAiGeneration(this.phClient, {
         ...posthogParams,
         model: geminiParams.model,
         provider: 'gemini',
@@ -2961,7 +2925,7 @@ class WrappedModels {
         output: formatResponseGemini(response),
         latency,
         baseURL: 'https://generativelanguage.googleapis.com',
-        params: params,
+        modelParameters: getModelParams(params),
         httpStatus: 200,
         usage: {
           inputTokens: metadata?.promptTokenCount ?? 0,
@@ -2977,8 +2941,7 @@ class WrappedModels {
       return response;
     } catch (error) {
       const latency = (Date.now() - startTime) / 1000;
-      const enrichedError = await sendEventWithErrorToPosthog({
-        client: this.phClient,
+      await captureAiGeneration(this.phClient, {
         ...posthogParams,
         model: geminiParams.model,
         provider: 'gemini',
@@ -2986,14 +2949,14 @@ class WrappedModels {
         output: [],
         latency,
         baseURL: 'https://generativelanguage.googleapis.com',
-        params: params,
+        modelParameters: getModelParams(params),
         usage: {
           inputTokens: 0,
           outputTokens: 0
         },
-        error: error
+        error
       });
-      throw enrichedError;
+      throw error;
     }
   }
   async *generateContentStream(params) {
@@ -3092,8 +3055,7 @@ class WrappedModels {
         role: 'assistant',
         content: accumulatedContent
       }] : [];
-      await sendEventToPosthog({
-        client: this.phClient,
+      await captureAiGeneration(this.phClient, {
         ...posthogParams,
         model: geminiParams.model,
         provider: 'gemini',
@@ -3102,7 +3064,7 @@ class WrappedModels {
         latency,
         timeToFirstToken,
         baseURL: 'https://generativelanguage.googleapis.com',
-        params: params,
+        modelParameters: getModelParams(params),
         httpStatus: 200,
         usage: {
           ...usage,
@@ -3114,8 +3076,7 @@ class WrappedModels {
       });
     } catch (error) {
       const latency = (Date.now() - startTime) / 1000;
-      const enrichedError = await sendEventWithErrorToPosthog({
-        client: this.phClient,
+      await captureAiGeneration(this.phClient, {
         ...posthogParams,
         model: geminiParams.model,
         provider: 'gemini',
@@ -3123,14 +3084,14 @@ class WrappedModels {
         output: [],
         latency,
         baseURL: 'https://generativelanguage.googleapis.com',
-        params: params,
+        modelParameters: getModelParams(params),
         usage: {
           inputTokens: 0,
           outputTokens: 0
         },
-        error: error
+        error
       });
-      throw enrichedError;
+      throw error;
     }
   }
   async embedContent(params) {
@@ -3143,8 +3104,7 @@ class WrappedModels {
       const response = await this.client.models.embedContent(geminiParams);
       const latency = (Date.now() - startTime) / 1000;
       const inputTokens = extractEmbeddingTokenCount(response);
-      await sendEventToPosthog({
-        client: this.phClient,
+      await captureAiGeneration(this.phClient, {
         ...posthogParams,
         eventType: AIEvent.Embedding,
         model: geminiParams.model,
@@ -3153,7 +3113,7 @@ class WrappedModels {
         output: null,
         latency,
         baseURL: 'https://generativelanguage.googleapis.com',
-        params: params,
+        modelParameters: getModelParams(params),
         httpStatus: 200,
         usage: {
           inputTokens
@@ -3162,8 +3122,7 @@ class WrappedModels {
       return response;
     } catch (error) {
       const latency = (Date.now() - startTime) / 1000;
-      const enrichedError = await sendEventWithErrorToPosthog({
-        client: this.phClient,
+      await captureAiGeneration(this.phClient, {
         ...posthogParams,
         eventType: AIEvent.Embedding,
         model: geminiParams.model,
@@ -3172,13 +3131,13 @@ class WrappedModels {
         output: null,
         latency,
         baseURL: 'https://generativelanguage.googleapis.com',
-        params: params,
+        modelParameters: getModelParams(params),
         usage: {
           inputTokens: 0
         },
-        error: error
+        error
       });
-      throw enrichedError;
+      throw error;
     }
   }
   formatPartsAsContentBlocks(parts) {
@@ -4714,5 +4673,5 @@ class Prompts {
   }
 }
 
-export { PostHogAnthropic as Anthropic, PostHogAzureOpenAI as AzureOpenAI, PostHogGoogleGenAI as GoogleGenAI, LangChainCallbackHandler, PostHogOpenAI as OpenAI, Prompts, wrapVercelLanguageModel as withTracing };
+export { AIEvent, PostHogAnthropic as Anthropic, PostHogAzureOpenAI as AzureOpenAI, PostHogGoogleGenAI as GoogleGenAI, LangChainCallbackHandler, PostHogOpenAI as OpenAI, Prompts, captureAiGeneration, wrapVercelLanguageModel as withTracing };
 //# sourceMappingURL=index.mjs.map