@posthog/ai 3.2.1 → 3.3.1

package/CHANGELOG.md

@@ -1,3 +1,13 @@
+# 3.3.1 - 2025-03-13
+
+- fix: fix vercel output mapping and token caching
+
+# 3.3.0 - 2025-03-08
+
+- feat: add reasoning and cache tokens to openai and anthropic
+- feat: add tool support for vercel
+- feat: add support for other media types vercel
+
 # 3.2.1 - 2025-02-11
 
 - fix: add experimental_wrapLanguageModel to vercel middleware supporting older versions of ai

package/lib/index.cjs.js

@@ -100,7 +100,8 @@ const sendEventToPosthog = ({
   httpStatus = 200,
   usage = {},
   isError = false,
-  error
+  error,
+  tools
 }) => {
   if (client.capture) {
     let errorData = {};
@@ -120,6 +121,17 @@ const sendEventToPosthog = ({
         $ai_total_cost_usd: inputCostUSD + outputCostUSD
       };
     }
+    let additionalTokenValues = {
+      ...(usage.reasoningTokens ? {
+        $ai_reasoning_tokens: usage.reasoningTokens
+      } : {}),
+      ...(usage.cacheReadInputTokens ? {
+        $ai_cache_read_input_tokens: usage.cacheReadInputTokens
+      } : {}),
+      ...(usage.cacheCreationInputTokens ? {
+        $ai_cache_creation_input_tokens: usage.cacheCreationInputTokens
+      } : {})
+    };
     client.capture({
       distinctId: distinctId ?? traceId,
       event: '$ai_generation',
@@ -132,6 +144,7 @@ const sendEventToPosthog = ({
         $ai_http_status: httpStatus,
         $ai_input_tokens: usage.inputTokens ?? 0,
         $ai_output_tokens: usage.outputTokens ?? 0,
+        ...additionalTokenValues,
         $ai_latency: latency,
         $ai_trace_id: traceId,
         $ai_base_url: baseURL,
@@ -139,6 +152,9 @@ const sendEventToPosthog = ({
         ...(distinctId ? {} : {
           $process_person_profile: false
         }),
+        ...(tools ? {
+          $ai_tools: tools
+        } : {}),
         ...errorData,
         ...costOverrideData
       },
@@ -200,7 +216,9 @@ class WrappedCompletions$1 extends OpenAIOrignal__default["default"].Chat.Comple
                 if (chunk.usage) {
                   usage = {
                     inputTokens: chunk.usage.prompt_tokens ?? 0,
-                    outputTokens: chunk.usage.completion_tokens ?? 0
+                    outputTokens: chunk.usage.completion_tokens ?? 0,
+                    reasoningTokens: chunk.usage.completion_tokens_details?.reasoning_tokens ?? 0,
+                    cacheReadInputTokens: chunk.usage.prompt_tokens_details?.cached_tokens ?? 0
                   };
                 }
               }
@@ -267,7 +285,9 @@ class WrappedCompletions$1 extends OpenAIOrignal__default["default"].Chat.Comple
             httpStatus: 200,
             usage: {
               inputTokens: result.usage?.prompt_tokens ?? 0,
-              outputTokens: result.usage?.completion_tokens ?? 0
+              outputTokens: result.usage?.completion_tokens ?? 0,
+              reasoningTokens: result.usage?.completion_tokens_details?.reasoning_tokens ?? 0,
+              cacheReadInputTokens: result.usage?.prompt_tokens_details?.cached_tokens ?? 0
             }
           });
         }
@@ -356,7 +376,9 @@ class WrappedCompletions extends OpenAIOrignal.AzureOpenAI.Chat.Completions {
                   }
                   usage = {
                     inputTokens: chunk.usage.prompt_tokens ?? 0,
-                    outputTokens: chunk.usage.completion_tokens ?? 0
+                    outputTokens: chunk.usage.completion_tokens ?? 0,
+                    reasoningTokens: chunk.usage.completion_tokens_details?.reasoning_tokens ?? 0,
+                    cacheReadInputTokens: chunk.usage.prompt_tokens_details?.cached_tokens ?? 0
                   };
                 }
               }
@@ -428,7 +450,9 @@ class WrappedCompletions extends OpenAIOrignal.AzureOpenAI.Chat.Completions {
             httpStatus: 200,
             usage: {
               inputTokens: result.usage?.prompt_tokens ?? 0,
-              outputTokens: result.usage?.completion_tokens ?? 0
+              outputTokens: result.usage?.completion_tokens ?? 0,
+              reasoningTokens: result.usage?.completion_tokens_details?.reasoning_tokens ?? 0,
+              cacheReadInputTokens: result.usage?.prompt_tokens_details?.cached_tokens ?? 0
             }
           });
         }
@@ -473,16 +497,60 @@ const mapVercelParams = params => {
 };
 const mapVercelPrompt = prompt => {
   return prompt.map(p => {
-    let content = '';
+    let content = {};
     if (Array.isArray(p.content)) {
       content = p.content.map(c => {
         if (c.type === 'text') {
-          return c.text;
+          return {
+            type: 'text',
+            content: c.text
+          };
+        } else if (c.type === 'image') {
+          return {
+            type: 'image',
+            content: {
+              // if image is a url use it, or use "none supported"
+              image: c.image instanceof URL ? c.image.toString() : 'raw images not supported',
+              mimeType: c.mimeType
+            }
+          };
+        } else if (c.type === 'file') {
+          return {
+            type: 'file',
+            content: {
+              file: c.data instanceof URL ? c.data.toString() : 'raw files not supported',
+              mimeType: c.mimeType
+            }
+          };
+        } else if (c.type === 'tool-call') {
+          return {
+            type: 'tool-call',
+            content: {
+              toolCallId: c.toolCallId,
+              toolName: c.toolName,
+              args: c.args
+            }
+          };
+        } else if (c.type === 'tool-result') {
+          return {
+            type: 'tool-result',
+            content: {
+              toolCallId: c.toolCallId,
+              toolName: c.toolName,
+              result: c.result,
+              isError: c.isError
+            }
+          };
         }
-        return '';
-      }).join('');
+        return {
+          content: ''
+        };
+      });
     } else {
-      content = p.content;
+      content = {
+        type: 'text',
+        text: p.content
+      };
     }
     return {
       role: p.role,
@@ -490,6 +558,45 @@ const mapVercelPrompt = prompt => {
     };
   });
 };
+const mapVercelOutput = result => {
+  let output = {
+    ...(result.text ? {
+      text: result.text
+    } : {}),
+    ...(result.object ? {
+      object: result.object
+    } : {}),
+    ...(result.reasoning ? {
+      reasoning: result.reasoning
+    } : {}),
+    ...(result.response ? {
+      response: result.response
+    } : {}),
+    ...(result.finishReason ? {
+      finishReason: result.finishReason
+    } : {}),
+    ...(result.usage ? {
+      usage: result.usage
+    } : {}),
+    ...(result.warnings ? {
+      warnings: result.warnings
+    } : {}),
+    ...(result.providerMetadata ? {
+      toolCalls: result.providerMetadata
+    } : {})
+  };
+  // if text and no object or reasoning, return text
+  if (output.text && !output.object && !output.reasoning) {
+    return [{
+      content: output.text,
+      role: 'assistant'
+    }];
+  }
+  return [{
+    content: JSON.stringify(output),
+    role: 'assistant'
+  }];
+};
 const extractProvider = model => {
   // vercel provider is in the format of provider.endpoint
   const provider = model.provider.toLowerCase();
@@ -513,11 +620,21 @@ const createInstrumentationMiddleware = (phClient, model, options) => {
         const modelId = options.posthogModelOverride ?? (result.response?.modelId ? result.response.modelId : model.modelId);
         const provider = options.posthogProviderOverride ?? extractProvider(model);
         const baseURL = ''; // cannot currently get baseURL from vercel
-        let content = result.text;
-        if (!content) {
-          // support generate Object
-          content = result.toolCalls?.[0].args || JSON.stringify(result);
-        }
+        let content = mapVercelOutput(result);
+        // let tools = result.toolCalls
+        let providerMetadata = result.providerMetadata;
+        let additionalTokenValues = {
+          ...(providerMetadata?.openai?.reasoningTokens ? {
+            reasoningTokens: providerMetadata.openai.reasoningTokens
+          } : {}),
+          ...(providerMetadata?.openai?.cachedPromptTokens ? {
+            cacheReadInputTokens: providerMetadata.openai.cachedPromptTokens
+          } : {}),
+          ...(providerMetadata?.anthropic ? {
+            cacheReadInputTokens: providerMetadata.anthropic.cacheReadInputTokens,
+            cacheCreationInputTokens: providerMetadata.anthropic.cacheCreationInputTokens
+          } : {})
+        };
         sendEventToPosthog({
           client: phClient,
           distinctId: options.posthogDistinctId,
@@ -535,7 +652,8 @@ const createInstrumentationMiddleware = (phClient, model, options) => {
           httpStatus: 200,
           usage: {
             inputTokens: result.usage.promptTokens,
-            outputTokens: result.usage.completionTokens
+            outputTokens: result.usage.completionTokens,
+            ...additionalTokenValues
           }
         });
         return result;
@@ -592,6 +710,18 @@ const createInstrumentationMiddleware = (phClient, model, options) => {
                 inputTokens: chunk.usage?.promptTokens,
                 outputTokens: chunk.usage?.completionTokens
               };
+              if (chunk.providerMetadata?.openai?.reasoningTokens) {
+                usage.reasoningTokens = chunk.providerMetadata.openai.reasoningTokens;
+              }
+              if (chunk.providerMetadata?.openai?.cachedPromptTokens) {
+                usage.cacheReadInputTokens = chunk.providerMetadata.openai.cachedPromptTokens;
+              }
+              if (chunk.providerMetadata?.anthropic?.cacheReadInputTokens) {
+                usage.cacheReadInputTokens = chunk.providerMetadata.anthropic.cacheReadInputTokens;
+              }
+              if (chunk.providerMetadata?.anthropic?.cacheCreationInputTokens) {
+                usage.cacheCreationInputTokens = chunk.providerMetadata.anthropic.cacheCreationInputTokens;
+              }
             }
             controller.enqueue(chunk);
           },
@@ -694,7 +824,9 @@ class WrappedMessages extends AnthropicOriginal__default["default"].Messages {
         let accumulatedContent = '';
         const usage = {
           inputTokens: 0,
-          outputTokens: 0
+          outputTokens: 0,
+          cacheCreationInputTokens: 0,
+          cacheReadInputTokens: 0
         };
         if ('tee' in value) {
           const [stream1, stream2] = value.tee();
@@ -709,6 +841,8 @@ class WrappedMessages extends AnthropicOriginal__default["default"].Messages {
                 }
                 if (chunk.type == 'message_start') {
                   usage.inputTokens = chunk.message.usage.input_tokens ?? 0;
+                  usage.cacheCreationInputTokens = chunk.message.usage.cache_creation_input_tokens ?? 0;
+                  usage.cacheReadInputTokens = chunk.message.usage.cache_read_input_tokens ?? 0;
                 }
                 if ('usage' in chunk) {
                   usage.outputTokens = chunk.usage.output_tokens ?? 0;
@@ -778,7 +912,9 @@ class WrappedMessages extends AnthropicOriginal__default["default"].Messages {
             httpStatus: 200,
             usage: {
               inputTokens: result.usage.input_tokens ?? 0,
-              outputTokens: result.usage.output_tokens ?? 0
+              outputTokens: result.usage.output_tokens ?? 0,
+              cacheCreationInputTokens: result.usage.cache_creation_input_tokens ?? 0,
+              cacheReadInputTokens: result.usage.cache_read_input_tokens ?? 0
             }
           });
         }
@@ -1549,6 +1685,9 @@ class LangChainCallbackHandler extends BaseCallbackHandler {
       $ai_latency: latency,
       $ai_base_url: run.baseUrl
     };
+    if (run.tools) {
+      eventProperties['$ai_tools'] = withPrivacyMode(this.client, this.privacyMode, run.tools);
+    }
     if (output instanceof Error) {
       eventProperties['$ai_http_status'] = output.status || 500;
       eventProperties['$ai_error'] = output.toString();