@posthog/ai 7.16.15 → 7.17.1

package/dist/vercel/index.mjs

@@ -1,8 +1,6 @@
 import { v4 } from 'uuid';
 import { uuidv7 } from '@posthog/core';
 
-var version = "7.16.15";
-
 // Type guards for safer type checking
 
 const isString = value => {
@@ -74,6 +72,14 @@ function getTokensSource(posthogProperties) {
 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
@@ -136,20 +142,16 @@ const truncate = input => {
   }
 
   // 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);
   }
@@ -292,73 +294,69 @@ function sanitizeValues(obj) {
   }
   return jsonSafe;
 }
-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, {
-      $ai_trace_id: traceId
-    }, exceptionId);
-    enrichedError.__posthog_previously_captured_error = true;
-    properties.exceptionId = exceptionId;
-  }
-  await sendEventToPosthog(properties);
-  return enrichedError;
-};
-const sendEventToPosthog = async ({
-  client,
-  eventType = AIEvent.Generation,
-  distinctId,
-  traceId,
-  model,
-  provider,
-  input,
-  output,
-  latency,
-  timeToFirstToken,
-  baseURL,
-  params,
-  httpStatus = 200,
-  usage = {},
-  error,
-  exceptionId,
-  stopReason,
-  tools,
-  captureImmediate = false
-}) => {
+
+var version = "7.17.1";
+
+/**
+ * Options for `captureAiGeneration`. Mirrors the `$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,
@@ -385,50 +383,48 @@ 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();
 };
 
 // Union types for dual version support
@@ -528,18 +524,26 @@ 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',
@@ -770,6 +774,19 @@ const wrapVercelLanguageModel = (model, phClient, options) => {
     }
   };
 
+  // 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, {
@@ -826,46 +843,40 @@ 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,
@@ -1017,10 +1028,8 @@ 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),
@@ -1028,12 +1037,11 @@ 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
               });
             }
           });
@@ -1042,26 +1050,23 @@ 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,