I’m new to Agent Builder and currently working on a multi-agent setup. My main agent is functioning correctly and can respond using the knowledge base I’ve provided.
However, I’m running into an issue when trying to add PDF or Word documents via File Search → Vectors / Files. Every time I attempt to upload files, I get a “failed” error, and the files are not added successfully.
I’ve tried multiple documents and formats, but I’m still unable to figure out what’s causing the issue or how to fix it.
Has anyone faced a similar problem or found a solution? Any guidance would be greatly appreciated.
Thanks in advance!
I haven’t personally hit this issue yet, but based on how the File Search / vector pipeline works — and looking at the screenshot — this doesn’t seem related to the agent setup itself. The vector store is created correctly, but the file is failing very early in the attach/vectorization step.
Even if you’ve already tried multiple documents and formats, these failures usually come from how the file is parsed and chunked before vectorization, not from the agent logic. A few structured things I’d suggest checking step by step:
• Start with a very small control file (1–2 pages, text-only, no images or tables) just to confirm the upload → vector flow works at all.
• Make sure PDFs contain real, selectable text. Scanned or image-based PDFs will fail unless they’ve been OCR’d.
• If the documents are large, split them into smaller parts. Size, heavy formatting, tables, or embedded images can silently break the vector step.
• For Word files like the one shown, try copying only the raw text into a brand-new document and re-saving it as a clean .docx, or export it to a simple text-based PDF. Older or heavily formatted Word files often cause early failures.
• If even a minimal file fails, it may point to a temporary platform issue, quota limit, or something still unstable in File Search since it’s evolving. Trying another browser or an incognito session can also help rule out UI issues.
The goal is to isolate whether this is document-related or pipeline-related. If a minimal file succeeds, you can gradually add complexity back until you see exactly what triggers the failure.
