[GH-ISSUE #2821] Improve Web Search with by preprocessing search query #28554

Closed
opened 2026-04-25 03:09:51 -05:00 by GiteaMirror · 0 comments
Owner

Originally created by @Expro on GitHub (Jun 4, 2024).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/2821

Is your feature request related to a problem? Please describe.

Currently, Web Search functionality simply forwards prompt that user put into chat toward search engine. This is not ideal, as chat encourages conversation-type expressions, while search engines prefers keyword-like expressions.

Describe the solution you'd like

It would be great if user prompt could be preprocessed before being being send to search engine. It could be as simple as allowing user to specify preprocessing system prompt for model that would be used for web search prepocessing.

Real world example:

Today I was testing Web Search functionality by asking LLama3 3 following prompt:

"What Azure services can be deployed on Azure HCI?"

This of course resulted in my SearXNG instance being queried with query "What Azure services can be deployed on Azure HCI?". Results were not bad, but they could be better.

I ran a simple experiment: in separate chat I've prompted LLama3 with following expression:

"Turn following text into search-friendly expression and answer with expression only, without quotation marks:

What Azure services can be deployed on Azure HCI?"

Llama 3 answered with: "azure hci supported services". Then I've used that answer as prompt with Web Search and received better answer.

Such solution would allow for natural, conversation-like prompts with search-friendly queries being used for web searches.

Originally created by @Expro on GitHub (Jun 4, 2024). Original GitHub issue: https://github.com/open-webui/open-webui/issues/2821 **Is your feature request related to a problem? Please describe.** Currently, Web Search functionality simply forwards prompt that user put into chat toward search engine. This is not ideal, as chat encourages conversation-type expressions, while search engines prefers keyword-like expressions. **Describe the solution you'd like** It would be great if user prompt could be preprocessed before being being send to search engine. It could be as simple as allowing user to specify preprocessing system prompt for model that would be used for web search prepocessing. Real world example: Today I was testing Web Search functionality by asking LLama3 3 following prompt: "What Azure services can be deployed on Azure HCI?" This of course resulted in my SearXNG instance being queried with query "What Azure services can be deployed on Azure HCI?". Results were not bad, but they could be better. I ran a simple experiment: in separate chat I've prompted LLama3 with following expression: "Turn following text into search-friendly expression and answer with expression only, without quotation marks: What Azure services can be deployed on Azure HCI?" Llama 3 answered with: "azure hci supported services". Then I've used that answer as prompt with Web Search and received better answer. Such solution would allow for natural, conversation-like prompts with search-friendly queries being used for web searches.
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/open-webui#28554