A document answer can miss a relevant passage even when the file is already indexed. Open WebUI RAG settings determine how much retrieved text reaches the chat model, and adjusting Top K offers a way to test a larger or smaller context limit.
With Hybrid Search off, Top K limits the number of retrieved chunks passed to the model. A higher limit can include more supporting text, but also more unrelated material and a larger share of the model's context window. With hybrid search enabled, Top K controls retrieval candidates and Top K Reranker limits the chunks retained afterward.
Document settings apply across the instance and require an administrator account. An already indexed knowledge base with a known answer provides a baseline for comparison; the chat model, question, documents, and retrieval mode must stay the same while one setting changes. Uploaded text remains stored on the instance, and retrieved passages go to the selected model provider.
Steps to tune RAG settings in Open WebUI:
- Send a known-answer question to the already attached Support Escalation Handbook before changing settings.
According to Support Escalation Handbook, what is the verification phrase in support-note.txt? Quote the phrase and cite the source.
The knowledge-base creation guide supplies the indexed support-note.txt and the attached collection. A unique phrase is easier to check than a general summary.
Related: How to create a knowledge base in Open WebUI - Confirm the baseline reply contains blue-sky-router with a source citation.

- Open your profile menu at the bottom of the sidebar.

- Select Settings.

- Scroll the settings navigation to the Documents entry under Admin.

- Select Documents.

- Record the current Top K value under Retrieval for rollback.
The starting value here is 3, with Hybrid Search and Full Context Mode off. The Bypass Embedding and Retrieval switch is also off. Full-context and bypass modes do not test the chunk limit.
Saving a new Top K value changes the retrieval limit for the instance. The value 4 is a small comparison change, not a universal recommendation.
- Change Top K from 3 to 4.
All other settings remain at their baseline values.
- Click Save to apply the retrieval setting.
A Top K change applies to subsequent queries without reindexing. Chunk size, overlap, and embedding-model changes affect indexing and require a separate rebuild of existing knowledge; the Reindex and Reset controls are unnecessary for this change.
- Click Back in the upper-left corner to leave Settings.

- Start a New Chat.

- Select the same chat model used for the baseline from the message box's model selector.

- Type # in the message box.

- Select Support Escalation Handbook under Collections.
The collection attachment uses Focused Retrieval by default. Full Context sends the whole document and would bypass the Top K comparison.
- Enter the same question used for the baseline.
According to Support Escalation Handbook, what is the verification phrase in support-note.txt? Quote the phrase and cite the source.
- Send the message.

- Check the reply for blue-sky-router.
A tiny note may contain only one chunk, so this check establishes that retrieval still works after the setting change. It does not establish that 4 improves ranking or that every question benefits from more context.
- Expand the source list beneath the reply.

- Open the cited support-note.txt passage.
The source should still contain weekdays 09:00-17:00 UTC, Platform Support Team, and blue-sky-router. The passage provides the comparison when a reply abbreviates or omits part of a value. If the expected passage disappears or answers worsen, restoring the recorded Top K value through Documents and Save returns the previous retrieval limit.