From 73b0b3149d90062fbbfbc5cea0da05a8e05f48d3 Mon Sep 17 00:00:00 2001
From: Rob Colbert <rob@digitaltelepresence.com>
Date: Mon, 18 May 2026 15:29:59 -0400
Subject: [PATCH] fix(ai): log raw OpenAI API response objects, not cooked
 usage

- Add rawUsage field to OpenAiChatIterationResult to carry the
  original OpenAI SDK usage object through the call chain
- All downstream chat() logs now dump rawUsage via
  JSON.parse(JSON.stringify()) instead of our cooked
  { promptTokens, completionTokens }
- This exposes completion_tokens_details.reasoning_tokens and
  any other fields OpenAI returns that we weren't capturing
---
 packages/ai/src/openai.ts | 19 ++++++++++++++-----
 1 file changed, 14 insertions(+), 5 deletions(-)

diff --git a/packages/ai/src/openai.ts b/packages/ai/src/openai.ts
index d8034f5..027baac 100644
--- a/packages/ai/src/openai.ts
+++ b/packages/ai/src/openai.ts
@@ -85,6 +85,8 @@ interface OpenAiChatIterationResult {
     promptTokens: number;
     completionTokens: number;
   };
+  // Raw usage object from OpenAI API response (for logging/observability)
+  rawUsage?: unknown;
 }
 
 export class OpenAiApi extends AiApi {
@@ -230,6 +232,7 @@ export class OpenAiApi extends AiApi {
     let accumulatedResponse = "";
     let accumulatedThinking = "";
     let usage: { promptTokens: number; completionTokens: number } | undefined;
+    let rawUsage: unknown;
 
     for await (const chunk of response) {
       if (options.signal?.aborted) {
@@ -242,6 +245,7 @@ export class OpenAiApi extends AiApi {
           promptTokens: chunk.usage.prompt_tokens,
           completionTokens: chunk.usage.completion_tokens,
         };
+        rawUsage = chunk.usage;
         // LOG: dump entire usage chunk from OpenAI generate streaming response
         this.log.info("OpenAI generate: streaming chunk with usage", {
           usage: JSON.parse(JSON.stringify(chunk.usage)),
@@ -277,7 +281,7 @@ export class OpenAiApi extends AiApi {
 
     // LOG: dump final usage and accumulated stats from generate()
     this.log.info("OpenAI generate: complete", {
-      usage,
+      rawUsage: rawUsage ? JSON.parse(JSON.stringify(rawUsage)) : undefined,
       responseLength: accumulatedResponse.length,
       thinkingLength: accumulatedThinking.length,
       durationMs,
@@ -335,7 +339,7 @@ export class OpenAiApi extends AiApi {
 
     // LOG: dump usage from streaming chat iteration
     this.log.info("OpenAI chat: streaming iteration usage", {
-      usage: iteration.usage,
+      rawUsage: iteration.rawUsage ? JSON.parse(JSON.stringify(iteration.rawUsage)) : undefined,
       finishReason: iteration.finishReason,
     });
 
@@ -352,7 +356,7 @@ export class OpenAiApi extends AiApi {
       });
       // LOG: dump usage from non-streaming fallback
       this.log.info("OpenAI chat: non-streaming fallback usage", {
-        usage: iteration.usage,
+        rawUsage: iteration.rawUsage ? JSON.parse(JSON.stringify(iteration.rawUsage)) : undefined,
         finishReason: iteration.finishReason,
       });
     }
@@ -377,7 +381,7 @@ export class OpenAiApi extends AiApi {
     };
     // LOG: dump final chat response stats
     this.log.info("OpenAI chat: final response", {
-      usage: iteration.usage,
+      rawUsage: iteration.rawUsage ? JSON.parse(JSON.stringify(iteration.rawUsage)) : undefined,
       stats: finalResponse.stats,
       finishReason: iteration.finishReason,
     });
@@ -470,6 +474,7 @@ export class OpenAiApi extends AiApi {
     let finishReason: string | null | undefined;
     const toolCallMap = new Map<number, StreamingToolCallAccumulator>();
     let usage: { promptTokens: number; completionTokens: number } | undefined;
+    let rawUsage: unknown;
 
     for await (const chunk of response) {
       if (signal?.aborted) {
@@ -485,6 +490,7 @@ export class OpenAiApi extends AiApi {
           promptTokens: chunk.usage.prompt_tokens,
           completionTokens: chunk.usage.completion_tokens,
         };
+        rawUsage = chunk.usage;
         // LOG: dump entire usage chunk from OpenAI chat streaming response
         this.log.info("OpenAI chat stream: streaming chunk with usage", {
           usage: JSON.parse(JSON.stringify(chunk.usage)),
@@ -522,7 +528,7 @@ export class OpenAiApi extends AiApi {
       contentDeltaCount,
       toolDeltaCount,
       finishReason,
-      usage,
+      rawUsage: rawUsage ? JSON.parse(JSON.stringify(rawUsage)) : undefined,
       responseLength: content.length,
       thinkingLength: thinking.length,
       toolCallCount: toolCallMap.size,
@@ -537,6 +543,7 @@ export class OpenAiApi extends AiApi {
       contentDeltaCount,
       toolDeltaCount,
       usage,
+      rawUsage,
     };
   }
 
@@ -580,6 +587,7 @@ export class OpenAiApi extends AiApi {
       promptTokens: response.usage.prompt_tokens,
       completionTokens: response.usage.completion_tokens,
     } : undefined;
+    const rawUsage = response.usage;
     const assistantToolCalls = (message?.tool_calls || [])
       .filter((tc) => tc.type === "function")
       .map((tc) => ({
@@ -607,6 +615,7 @@ export class OpenAiApi extends AiApi {
       contentDeltaCount: content ? 1 : 0,
       toolDeltaCount: assistantToolCalls.length,
       usage,
+      rawUsage,
     };
   }