@posthog/ai 7.3.1 → 7.4.0

package/dist/index.mjs

@@ -2,10 +2,11 @@ import { OpenAI, AzureOpenAI } from 'openai';
 import { Buffer } from 'buffer';
 import * as uuid from 'uuid';
 import { v4 } from 'uuid';
+import { uuidv7 } from '@posthog/core';
 import AnthropicOriginal from '@anthropic-ai/sdk';
 import { GoogleGenAI } from '@google/genai';
 
-var version = "7.3.1";
+var version = "7.4.0";
 
 // Type guards for safer type checking
 const isString = value => {
@@ -473,6 +474,33 @@ 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,
@@ -487,8 +515,8 @@ const sendEventToPosthog = async ({
   params,
   httpStatus = 200,
   usage = {},
-  isError = false,
   error,
+  exceptionId,
   tools,
   captureImmediate = false
 }) => {
@@ -500,10 +528,11 @@ const sendEventToPosthog = async ({
   const safeOutput = sanitizeValues(output);
   const safeError = sanitizeValues(error);
   let errorData = {};
-  if (isError) {
+  if (error) {
     errorData = {
       $ai_is_error: true,
-      $ai_error: safeError
+      $ai_error: safeError,
+      $exception_event_id: exceptionId
     };
   }
   let costOverrideData = {};
@@ -569,6 +598,7 @@ const sendEventToPosthog = async ({
   } else {
     client.capture(event);
   }
+  return Promise.resolve();
 };
 function formatOpenAIResponsesInput(input, instructions) {
   const messages = [];
@@ -976,8 +1006,7 @@ let WrappedCompletions$1 = class WrappedCompletions extends Completions {
                 tools: availableTools
               });
             } catch (error) {
-              const httpStatus = error && typeof error === 'object' && 'status' in error ? error.status ?? 500 : 500;
-              await sendEventToPosthog({
+              const enrichedError = await sendEventWithErrorToPosthog({
                 client: this.phClient,
                 ...posthogParams,
                 model: openAIParams.model,
@@ -987,14 +1016,13 @@ let WrappedCompletions$1 = class WrappedCompletions extends Completions {
                 latency: 0,
                 baseURL: this.baseURL,
                 params: body,
-                httpStatus,
                 usage: {
                   inputTokens: 0,
                   outputTokens: 0
                 },
-                isError: true,
-                error: JSON.stringify(error)
+                error
               });
+              throw enrichedError;
             }
           })();
           // Return the other stream to the user
@@ -1047,7 +1075,6 @@ let WrappedCompletions$1 = class WrappedCompletions extends Completions {
             inputTokens: 0,
             outputTokens: 0
           },
-          isError: true,
           error: JSON.stringify(error)
         });
         throw error;
@@ -1130,8 +1157,7 @@ let WrappedResponses$1 = class WrappedResponses extends Responses {
                 tools: availableTools
               });
             } catch (error) {
-              const httpStatus = error && typeof error === 'object' && 'status' in error ? error.status ?? 500 : 500;
-              await sendEventToPosthog({
+              const enrichedError = await sendEventWithErrorToPosthog({
                 client: this.phClient,
                 ...posthogParams,
                 model: openAIParams.model,
@@ -1141,14 +1167,13 @@ let WrappedResponses$1 = class WrappedResponses extends Responses {
                 latency: 0,
                 baseURL: this.baseURL,
                 params: body,
-                httpStatus,
                 usage: {
                   inputTokens: 0,
                   outputTokens: 0
                 },
-                isError: true,
-                error: JSON.stringify(error)
+                error: error
               });
+              throw enrichedError;
             }
           })();
           return stream2;
@@ -1202,7 +1227,6 @@ let WrappedResponses$1 = class WrappedResponses extends Responses {
             inputTokens: 0,
             outputTokens: 0
           },
-          isError: true,
           error: JSON.stringify(error)
         });
         throw error;
@@ -1244,8 +1268,7 @@ 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({
+        const enrichedError = await sendEventWithErrorToPosthog({
           client: this.phClient,
           ...posthogParams,
           model: openAIParams.model,
@@ -1255,15 +1278,13 @@ let WrappedResponses$1 = class WrappedResponses extends Responses {
           latency: 0,
           baseURL: this.baseURL,
           params: body,
-          httpStatus,
           usage: {
             inputTokens: 0,
             outputTokens: 0
           },
-          isError: true,
           error: JSON.stringify(error)
         });
-        throw error;
+        throw enrichedError;
       });
       return wrappedPromise;
     } finally {
@@ -1323,7 +1344,6 @@ let WrappedEmbeddings$1 = class WrappedEmbeddings extends Embeddings {
         usage: {
           inputTokens: 0
         },
-        isError: true,
         error: JSON.stringify(error)
       });
       throw error;
@@ -1391,8 +1411,7 @@ class WrappedTranscriptions extends Transcriptions {
                 tools: availableTools
               });
             } catch (error) {
-              const httpStatus = error && typeof error === 'object' && 'status' in error ? error.status ?? 500 : 500;
-              await sendEventToPosthog({
+              const enrichedError = await sendEventWithErrorToPosthog({
                 client: this.phClient,
                 ...posthogParams,
                 model: openAIParams.model,
@@ -1402,14 +1421,13 @@ class WrappedTranscriptions extends Transcriptions {
                 latency: 0,
                 baseURL: this.baseURL,
                 params: body,
-                httpStatus,
                 usage: {
                   inputTokens: 0,
                   outputTokens: 0
                 },
-                isError: true,
-                error: JSON.stringify(error)
+                error: error
               });
+              throw enrichedError;
             }
           })();
           return stream2;
@@ -1439,8 +1457,7 @@ class WrappedTranscriptions extends Transcriptions {
           return result;
         }
       }, async error => {
-        const httpStatus = error && typeof error === 'object' && 'status' in error ? error.status ?? 500 : 500;
-        await sendEventToPosthog({
+        const enrichedError = await sendEventWithErrorToPosthog({
           client: this.phClient,
           ...posthogParams,
           model: openAIParams.model,
@@ -1450,15 +1467,13 @@ class WrappedTranscriptions extends Transcriptions {
           latency: 0,
           baseURL: this.baseURL,
           params: body,
-          httpStatus,
           usage: {
             inputTokens: 0,
             outputTokens: 0
           },
-          isError: true,
-          error: JSON.stringify(error)
+          error: error
         });
-        throw error;
+        throw enrichedError;
       });
       return wrappedPromise;
     }
@@ -1609,8 +1624,7 @@ class WrappedCompletions extends AzureOpenAI.Chat.Completions {
                 usage
               });
             } catch (error) {
-              const httpStatus = error && typeof error === 'object' && 'status' in error ? error.status ?? 500 : 500;
-              await sendEventToPosthog({
+              const enrichedError = await sendEventWithErrorToPosthog({
                 client: this.phClient,
                 ...posthogParams,
                 model: openAIParams.model,
@@ -1620,14 +1634,13 @@ class WrappedCompletions extends AzureOpenAI.Chat.Completions {
                 latency: 0,
                 baseURL: this.baseURL,
                 params: body,
-                httpStatus,
                 usage: {
                   inputTokens: 0,
                   outputTokens: 0
                 },
-                isError: true,
-                error: JSON.stringify(error)
+                error: error
               });
+              throw enrichedError;
             }
           })();
           // Return the other stream to the user
@@ -1676,7 +1689,6 @@ class WrappedCompletions extends AzureOpenAI.Chat.Completions {
             inputTokens: 0,
             outputTokens: 0
           },
-          isError: true,
           error: JSON.stringify(error)
         });
         throw error;
@@ -1745,8 +1757,7 @@ class WrappedResponses extends AzureOpenAI.Responses {
                 usage
               });
             } catch (error) {
-              const httpStatus = error && typeof error === 'object' && 'status' in error ? error.status ?? 500 : 500;
-              await sendEventToPosthog({
+              const enrichedError = await sendEventWithErrorToPosthog({
                 client: this.phClient,
                 ...posthogParams,
                 model: openAIParams.model,
@@ -1756,14 +1767,13 @@ class WrappedResponses extends AzureOpenAI.Responses {
                 latency: 0,
                 baseURL: this.baseURL,
                 params: body,
-                httpStatus,
                 usage: {
                   inputTokens: 0,
                   outputTokens: 0
                 },
-                isError: true,
-                error: JSON.stringify(error)
+                error: error
               });
+              throw enrichedError;
             }
           })();
           return stream2;
@@ -1811,7 +1821,6 @@ class WrappedResponses extends AzureOpenAI.Responses {
             inputTokens: 0,
             outputTokens: 0
           },
-          isError: true,
           error: JSON.stringify(error)
         });
         throw error;
@@ -1863,7 +1872,6 @@ class WrappedResponses extends AzureOpenAI.Responses {
           inputTokens: 0,
           outputTokens: 0
         },
-        isError: true,
         error: JSON.stringify(error)
       });
       throw error;
@@ -1921,7 +1929,6 @@ class WrappedEmbeddings extends AzureOpenAI.Embeddings {
         usage: {
           inputTokens: 0
         },
-        isError: true,
         error: JSON.stringify(error)
       });
       throw error;
@@ -2249,7 +2256,7 @@ const wrapVercelLanguageModel = (model, phClient, options) => {
         return result;
       } catch (error) {
         const modelId = model.modelId;
-        await sendEventToPosthog({
+        const enrichedError = await sendEventWithErrorToPosthog({
           client: phClient,
           distinctId: mergedOptions.posthogDistinctId,
           traceId: mergedOptions.posthogTraceId ?? v4(),
@@ -2260,17 +2267,15 @@ const wrapVercelLanguageModel = (model, phClient, options) => {
           latency: 0,
           baseURL: '',
           params: mergedParams,
-          httpStatus: error?.status ? error.status : 500,
           usage: {
             inputTokens: 0,
             outputTokens: 0
           },
-          isError: true,
-          error: truncate(JSON.stringify(error)),
+          error: error,
           tools: availableTools,
           captureImmediate: mergedOptions.posthogCaptureImmediate
         });
-        throw error;
+        throw enrichedError;
       }
     },
     doStream: async params => {
@@ -2407,7 +2412,7 @@ const wrapVercelLanguageModel = (model, phClient, options) => {
           ...rest
         };
       } catch (error) {
-        await sendEventToPosthog({
+        const enrichedError = await sendEventWithErrorToPosthog({
           client: phClient,
           distinctId: mergedOptions.posthogDistinctId,
           traceId: mergedOptions.posthogTraceId ?? v4(),
@@ -2418,17 +2423,15 @@ const wrapVercelLanguageModel = (model, phClient, options) => {
           latency: 0,
           baseURL: '',
           params: mergedParams,
-          httpStatus: error?.status ? error.status : 500,
           usage: {
             inputTokens: 0,
             outputTokens: 0
           },
-          isError: true,
-          error: truncate(JSON.stringify(error)),
+          error: error,
           tools: availableTools,
           captureImmediate: mergedOptions.posthogCaptureImmediate
         });
-        throw error;
+        throw enrichedError;
       }
     }
   };
@@ -2585,8 +2588,7 @@ class WrappedMessages extends AnthropicOriginal.Messages {
                 tools: availableTools
               });
             } catch (error) {
-              // error handling
-              await sendEventToPosthog({
+              const enrichedError = await sendEventWithErrorToPosthog({
                 client: this.phClient,
                 ...posthogParams,
                 model: anthropicParams.model,
@@ -2596,14 +2598,13 @@ class WrappedMessages extends AnthropicOriginal.Messages {
                 latency: 0,
                 baseURL: this.baseURL,
                 params: body,
-                httpStatus: error?.status ? error.status : 500,
                 usage: {
                   inputTokens: 0,
                   outputTokens: 0
                 },
-                isError: true,
-                error: JSON.stringify(error)
+                error: error
               });
+              throw enrichedError;
             }
           })();
           // Return the other stream to the user
@@ -2654,7 +2655,6 @@ class WrappedMessages extends AnthropicOriginal.Messages {
             inputTokens: 0,
             outputTokens: 0
           },
-          isError: true,
           error: JSON.stringify(error)
         });
         throw error;
@@ -2714,7 +2714,7 @@ class WrappedModels {
       return response;
     } catch (error) {
       const latency = (Date.now() - startTime) / 1000;
-      await sendEventToPosthog({
+      const enrichedError = await sendEventWithErrorToPosthog({
         client: this.phClient,
         ...posthogParams,
         model: geminiParams.model,
@@ -2724,15 +2724,13 @@ class WrappedModels {
         latency,
         baseURL: 'https://generativelanguage.googleapis.com',
         params: params,
-        httpStatus: error?.status ?? 500,
         usage: {
           inputTokens: 0,
           outputTokens: 0
         },
-        isError: true,
-        error: JSON.stringify(error)
+        error: error
       });
-      throw error;
+      throw enrichedError;
     }
   }
   async *generateContentStream(params) {
@@ -2834,7 +2832,7 @@ class WrappedModels {
       });
     } catch (error) {
       const latency = (Date.now() - startTime) / 1000;
-      await sendEventToPosthog({
+      const enrichedError = await sendEventWithErrorToPosthog({
         client: this.phClient,
         ...posthogParams,
         model: geminiParams.model,
@@ -2844,15 +2842,13 @@ class WrappedModels {
         latency,
         baseURL: 'https://generativelanguage.googleapis.com',
         params: params,
-        httpStatus: error?.status ?? 500,
         usage: {
           inputTokens: 0,
           outputTokens: 0
         },
-        isError: true,
-        error: JSON.stringify(error)
+        error: error
       });
-      throw error;
+      throw enrichedError;
     }
   }
   formatPartsAsContentBlocks(parts) {
@@ -3232,6 +3228,64 @@ function mapKeys(fields, mapper, map) {
 	return mapped;
 }
 
+//#region src/load/validation.ts
+/**
+* Sentinel key used to mark escaped user objects during serialization.
+*
+* When a plain object contains 'lc' key (which could be confused with LC objects),
+* we wrap it as `{"__lc_escaped__": {...original...}}`.
+*/
+const LC_ESCAPED_KEY = "__lc_escaped__";
+/**
+* Check if an object needs escaping to prevent confusion with LC objects.
+*
+* An object needs escaping if:
+* 1. It has an `'lc'` key (could be confused with LC serialization format)
+* 2. It has only the escape key (would be mistaken for an escaped object)
+*/
+function needsEscaping(obj) {
+	return "lc" in obj || Object.keys(obj).length === 1 && LC_ESCAPED_KEY in obj;
+}
+/**
+* Wrap an object in the escape marker.
+*
+* @example
+* ```typescript
+* {"key": "value"}  // becomes {"__lc_escaped__": {"key": "value"}}
+* ```
+*/
+function escapeObject(obj) {
+	return { [LC_ESCAPED_KEY]: obj };
+}
+/**
+* Check if an object looks like a Serializable instance (duck typing).
+*/
+function isSerializableLike(obj) {
+	return obj !== null && typeof obj === "object" && "lc_serializable" in obj && typeof obj.toJSON === "function";
+}
+/**
+* Escape a value if it needs escaping (contains `lc` key).
+*
+* This is a simpler version of `serializeValue` that doesn't handle Serializable
+* objects - it's meant to be called on kwargs values that have already been
+* processed by `toJSON()`.
+*
+* @param value - The value to potentially escape.
+* @returns The value with any `lc`-containing objects wrapped in escape markers.
+*/
+function escapeIfNeeded(value) {
+	if (value !== null && typeof value === "object" && !Array.isArray(value)) {
+		if (isSerializableLike(value)) return value;
+		const record = value;
+		if (needsEscaping(record)) return escapeObject(record);
+		const result = {};
+		for (const [key, val] of Object.entries(record)) result[key] = escapeIfNeeded(val);
+		return result;
+	}
+	if (Array.isArray(value)) return value.map((item) => escapeIfNeeded(item));
+	return value;
+}
+
 //#region src/load/serializable.ts
 var serializable_exports = {};
 __export(serializable_exports, {
@@ -3353,11 +3407,15 @@ var Serializable = class Serializable {
 			}
 			if (last in read && read[last] !== void 0) write[last] = write[last] || read[last];
 		});
+		const escapedKwargs = {};
+		for (const [key, value] of Object.entries(kwargs)) escapedKwargs[key] = escapeIfNeeded(value);
+		const kwargsWithSecrets = Object.keys(secrets).length ? replaceSecrets(escapedKwargs, secrets) : escapedKwargs;
+		const processedKwargs = mapKeys(kwargsWithSecrets, keyToJson, aliases);
 		return {
 			lc: 1,
 			type: "constructor",
 			id: this.lc_id,
-			kwargs: mapKeys(Object.keys(secrets).length ? replaceSecrets(kwargs, secrets) : kwargs, keyToJson, aliases)
+			kwargs: processedKwargs
 		};
 	}
 	toJSONNotImplemented() {
@@ -3818,7 +3876,10 @@ class LangChainCallbackHandler extends BaseCallbackHandler {
       eventProperties['$ai_output_tokens'] = outputTokens;
       // Add additional token data to properties
       if (additionalTokenData.cacheReadInputTokens) {
-        eventProperties['$ai_cache_read_tokens'] = additionalTokenData.cacheReadInputTokens;
+        eventProperties['$ai_cache_read_input_tokens'] = additionalTokenData.cacheReadInputTokens;
+      }
+      if (additionalTokenData.cacheWriteInputTokens) {
+        eventProperties['$ai_cache_creation_input_tokens'] = additionalTokenData.cacheWriteInputTokens;
       }
       if (additionalTokenData.reasoningTokens) {
         eventProperties['$ai_reasoning_tokens'] = additionalTokenData.reasoningTokens;
@@ -3982,6 +4043,14 @@ class LangChainCallbackHandler extends BaseCallbackHandler {
       additionalTokenData.cacheReadInputTokens = usage.input_token_details.cache_read;
     } else if (usage.cachedPromptTokens != null) {
       additionalTokenData.cacheReadInputTokens = usage.cachedPromptTokens;
+    } else if (usage.cache_read_input_tokens != null) {
+      additionalTokenData.cacheReadInputTokens = usage.cache_read_input_tokens;
+    }
+    // Check for cache write/creation tokens in various formats
+    if (usage.cache_creation_input_tokens != null) {
+      additionalTokenData.cacheWriteInputTokens = usage.cache_creation_input_tokens;
+    } else if (usage.input_token_details?.cache_creation != null) {
+      additionalTokenData.cacheWriteInputTokens = usage.input_token_details.cache_creation;
     }
     // Check for reasoning tokens in various formats
     if (usage.completion_tokens_details?.reasoning_tokens != null) {
@@ -4027,8 +4096,10 @@ class LangChainCallbackHandler extends BaseCallbackHandler {
     if (webSearchCount !== undefined) {
       additionalTokenData.webSearchCount = webSearchCount;
     }
-    // For Anthropic providers, LangChain reports input_tokens as the sum of input and cache read tokens.
+    // For Anthropic providers, LangChain reports input_tokens as the sum of all input tokens.
     // Our cost calculation expects them to be separate for Anthropic, so we subtract cache tokens.
+    // Both cache_read and cache_write tokens should be subtracted since Anthropic's raw API
+    // reports input_tokens as tokens NOT read from or used to create a cache.
     // For other providers (OpenAI, etc.), input_tokens already excludes cache tokens as expected.
     // Match logic consistent with plugin-server: exact match on provider OR substring match on model
     let isAnthropic = false;
@@ -4037,8 +4108,11 @@ class LangChainCallbackHandler extends BaseCallbackHandler {
     } else if (model && model.toLowerCase().includes('anthropic')) {
       isAnthropic = true;
     }
-    if (isAnthropic && parsedUsage.input && additionalTokenData.cacheReadInputTokens) {
-      parsedUsage.input = Math.max(parsedUsage.input - additionalTokenData.cacheReadInputTokens, 0);
+    if (isAnthropic && parsedUsage.input) {
+      const cacheTokens = (additionalTokenData.cacheReadInputTokens || 0) + (additionalTokenData.cacheWriteInputTokens || 0);
+      if (cacheTokens > 0) {
+        parsedUsage.input = Math.max(parsedUsage.input - cacheTokens, 0);
+      }
     }
     return [parsedUsage.input, parsedUsage.output, additionalTokenData];
   }