v0.6 openwebui - not able to make it working

[sir3mat]

Hi, i'm trying to reproduce the autogen rag example using gemma 3 with openwebui v0.6 but I'm not able to make it working. It stops the searches and always return empty []
Do you have any sugestions?

[andsty]

Me also it stacks at internet search. if i refresh browser it showing something but not the final result

[sir3mat]

I also tried with models on vllm (Gemma 3 and Mistral small 24B)
It seems not able to search online
Searxng is configured correctly and I am able to perform web search with OpenWebUI (with button). But using this pipe it seems not able to work

Also with a knowledge base passed in input it seems not able to work

[myrulezzz]

You modified script for vllm?

[sir3mat]

I tried with ollama and granite models, I'm not able to make it works
Then I tried with different models hosted with vllm, in order to use them with this code I only changed base_url hey and add api_key key in the Jsons used to set LLM config.
The requests are done, I can see the autogen cli output but websearch is not working and the pipe return that it is not able to respond to user query because the provided context is []

[kellyaa]

Hi! Thank you for opening this issue! You can use any search provider that open webui supports -- whether it's Searxng or any other search API service. (The nice thing about Searxng is that you can search without an API key, though it can be cumbersome to setup). Have you confirmed that you've setup connectivity to the search provider in the Open WebUI Admin Panel? Admin Panel -> Settings -> Web Search. I've attached a screenshot.

[sir3mat]

Yep, if I perform web search normally using the OpenWebUI button "web search" in the chat UI it works well (searxng is called correctly)
I also tried the pipe with OpenWebUI 0.6.2 with no success

[kellyaa]

Ah that's super strange. Would you be able to attach the open webui logs here for the full run of the agent?

[andsty]

EXECUTING FUNCTION web_search...
Call ID: ollama_func_659
Input arguments: {'search_instruction': '4G box eligibility criteria'}
User (to Research_Assistant):

***** Response from calling tool (ollama_func_659) *****
Error: Config key 'RAG_WEB_SEARCH_ENGINE' not found
i think the error is somewhere here. the agents keep going into a loop

[kellyaa]

Thank you! v0.6 changed the RAG_WEB_SEARCH_ENGINE to WEB_SEARCH_ENGINE - I've updated the code accordingly! Let me know if this works for you.

[andsty]

Ok now it seems it doesn't stack at the web search but the agents are going into a loop. never get an answer. if i restart openwebui and go again to the ui i get an answer. After sources it starts again with
For detailed instructions and specific commands, refer to Oracle's official documentation or tutorials such as:

Cloning Pluggable Databases
Cloning a PDB or Non-CDB Creating a plan...
Executing step: Step 1: Backup the source pluggable database.
Planning the next step...
Executing step: "Step 2: Transfer the backup to the target Oracle Database Appliance."
Planning the next step...
Executing step: "Transfer backup files from local system to Object Storage bucket for subsequent retrieval on target Oracle Database Appliance."
Planning the next step...
Executing step: "Restore RMAN Configuration for Object Storage Service"
Planning the next step...
Executing step: "Clone Pluggable DB from Object Storage to Target CDB"
Planning the next step...
Executing step: "Validate Cloned PDB's Integrity and Functionality"
Summing up findings...

[kellyaa]

Hi! If the answer doesn't show up after "summing up findings", try just a refresh of the browser. For some reason after running the agent code for a while the UI doesn't always refresh on its own when the final answer appears. I opened a discussion/issue on this in OpenWebUI some time ago, but it hasn't ever been fully solved...

[myrulezzz]

Hi.So for crewai example this is the issue.for autogen though this is the issue again plus that it seems agents gi unto a loop beefiee finalizing and when refreshing get the output

[kellyaa]

@myrulezzz If you have the open webui logs for this scenario, that would be super helpful!

[kellyaa]

Update: Results should now show up without a browser refresh. It seems the synchronous calls to the LLM were causing issues. Switching to async cleared up these issues: #31