[Bug] Web search source retrieve

[vitspvt]

Environment: Docker deploy

Test method: add code to file src\lib\searxng.ts add --> console.log('Request URL:', url.toString()); before: const res = await axios.get(url.toString());

Example case: Search: Samsung S25

In first time it always No Sources --> Example Domain and Request URL is not log to me (May be not search anything)

but if you do it again with same query in same page, it will get correct web search resources and log Request URL is http://searxng:8080/search?format=json&q=Samsung+S25&language=all&engines=

I think this is bug.

[ItzCrazyKns]

If you're using a small model it fails to understand the prompt and gives random links for searching, please use a slightly bigger/better model.

[zhudemiao]

I used a locally deployed model deepseek-r1:14b and encountered the same issue: ① 'sources' appeared for the first time (with search URLs), but did not appear the second time. Moreover, when it appeared, the search results were completely unrelated to what I requested (randomly appearing). ② The second time, 'sources' (without any search URLs) did not appear

[Twilek-de]

Seeing the same thing. It doesn´t seem to pull any searxng results. Has Searxng changed anything? (Running llama3.2, or is that too small?)

[mpstaton]

Ditto. It seems to be less of a bug and more of a glitch. The first answer in a conversation thread comes back with only cited sources as if it's an academic paper and not Markdown sources with hyperlinks.

Once the second and third questions are going it seems to come back with the Searxng results.

[ItzCrazyKns]

Smaller models tend to not understand the prompt correctly and start messing things up. Please try using a bigger model. A lot of users are facing problems related to this, I can change how search is handled and use function calling but that would limit the models that can be used within the app (to models that support tool/function calling).