RAG Returns No Results
If your agent seems to be ignoring its knowledge base, or a RAG Query workflow step returns empty results, one of these causes is likely to blame.
Step 1: Check the document's indexing status
Open the Documents library and find the document. Look at the status badge:
| Status | Meaning |
|---|---|
| Processing | Still being chunked and embedded — wait a few minutes |
| Indexed | Ready to use |
| Failed | Processing failed; the document cannot be queried |
If the status is Failed, try re-uploading the file. If it fails again, the file is likely in a format that cannot be parsed (for example, a scanned PDF with no OCR text). Try exporting it in a different format.
Step 2: Check that the RAG tool is enabled on the agent
For agents, the knowledge base is only queried if the RAG tool is enabled.
- Open the agent builder.
- Click the Tools tab.
- Confirm the RAG toggle is on.
If it was off, enable it, save, and test again.
Step 3: Check the document is actually attached to the agent
Having a document in your library does not automatically make it available to every agent. Documents must be explicitly attached in the agent's Knowledge tab.
- Open the agent builder → Knowledge tab.
- Confirm the document you want to query is listed there.
- If it's not, click Add documents and attach it.
For workflow RAG Query steps, there is no separate "attach" step — the step queries your entire document library by default, unless you configure a document filter in the step settings.
Step 4: The query may be too vague
RAG retrieves chunks that are semantically similar to your query. A vague query like "tell me about the product" has low specificity and may not match any specific chunk closely enough.
Fix: Be more specific. Instead of "tell me about the product", try "what are the system requirements for installing the software" or "what is the return policy for international orders".
Step 5: The document may have poor extraction quality
If the document is a scanned PDF without embedded text, the chunks are essentially empty — there is nothing for the embedding model to work with.
How to check: Download the document from the library and try to select text in a PDF viewer. If you cannot select any text, the file is image-only.
Fix: Re-create the PDF from the original source file (Word, Google Docs, etc.) and upload the text-based version instead.
Step 6: Use the Reindex button
If the document content has changed since it was first indexed, the embeddings may be stale.
- Go to the Documents library.
- Click the document to open its detail view.
- Click Reindex.
This rebuilds all chunks and embeddings from the current file content.
Reindexing clears existing embeddings and rebuilds from scratch. During the reindex process (which may take a few minutes for large documents), the document temporarily shows Processing status and cannot be queried.
Was this helpful?