Iterate page-by-page: Instead of failing on the whole document, it now processes each page individually.

Graceful Skip: If a specific page fails (like the one with the 'bbox' error), it logs a warning, skips that page, inserts a placeholder [Error extracting page X], and continues to the next page.

app.py

@@ -446,11 +446,21 @@ def query_agent(question: str, files: Optional[List[FilePayload]] = None) -> str
                             if fmt == 'pdf':
                                 try:
                                     pdf_reader = pypdf.PdfReader(io.BytesIO(file_bytes))
-                                    for page in pdf_reader.pages:
-                                        extracted_text += page.extract_text() + "\n"
+                                    for i, page in enumerate(pdf_reader.pages):
+                                        try:
+                                            text = page.extract_text()
+                                            if text:
+                                                extracted_text += text + "\n"
+                                        except Exception as page_err:
+                                            logger.warning(f"Failed to extract text from page {i} of {file_obj.name}: {page_err}")
+                                            extracted_text += f"[Error extracting page {i+1}]\n"
+
+                                    if not extracted_text.strip():
+                                         extracted_text = "[No text could be extracted from this PDF. It might be an image-only PDF.]"
+
                                 except Exception as pdf_err:
                                     logger.error(f"PDF extraction failed for {file_obj.name}: {pdf_err}")
-                                    extracted_text = f"[Error extracting PDF text for {file_obj.name}]"
+                                    extracted_text = f"[Error extracting PDF text for {file_obj.name}: {str(pdf_err)}]"
                             else:
                                 # Try decoding as plain text (csv, txt, md, html, etc)
                                 try: