From 86a156db2627d68fb00b320c9ef7ca508a2a0df9 Mon Sep 17 00:00:00 2001
From: Vignesh Natarajan <vigneshnatarajan92@gmail.com>
Date: Sat, 14 Feb 2026 19:22:15 -0800
Subject: [PATCH] fix (media-understanding): treat binary application mimes as
 non-text

---
 src/media-understanding/apply.ts | 26 +++++++++++++++++++++++++-
 1 file changed, 25 insertions(+), 1 deletion(-)

diff --git a/src/media-understanding/apply.ts b/src/media-understanding/apply.ts
index 766549afc..2edf6ac5e 100644
--- a/src/media-understanding/apply.ts
+++ b/src/media-understanding/apply.ts
@@ -321,7 +321,31 @@ function isBinaryMediaMime(mime?: string): boolean {
   if (!mime) {
     return false;
   }
-  return mime.startsWith("image/") || mime.startsWith("audio/") || mime.startsWith("video/");
+  if (mime.startsWith("image/") || mime.startsWith("audio/") || mime.startsWith("video/")) {
+    return true;
+  }
+  if (mime === "application/octet-stream") {
+    return true;
+  }
+  if (
+    mime === "application/zip" ||
+    mime === "application/x-zip-compressed" ||
+    mime === "application/gzip" ||
+    mime === "application/x-gzip" ||
+    mime === "application/x-rar-compressed" ||
+    mime === "application/x-7z-compressed"
+  ) {
+    return true;
+  }
+  if (mime.startsWith("application/vnd.")) {
+    // Keep vendor +json/+xml payloads eligible for text extraction while
+    // treating the common binary vendor family (Office, archives, etc.) as binary.
+    if (mime.endsWith("+json") || mime.endsWith("+xml")) {
+      return false;
+    }
+    return true;
+  }
+  return false;
 }
 
 async function extractFileBlocks(params: {