I have searched the existing issues and discussions.
I am using the latest version of Open WebUI.
Installation Method
Pip Install
Open WebUI Version
0.6.5
Ollama Version (if applicable)
0.6.7
Operating System
Ubuntu 24.04
Browser (if applicable)
No response
Confirmation
I have read and followed all instructions in README.md.
I am using the latest version of both Open WebUI and Ollama.
I have included the browser console logs.
I have included the Docker container logs.
I have listed steps to reproduce the bug in detail.
Expected Behavior
Enabling "Web Search" should allow a model to search the web and use web results when formulating its response.
Actual Behavior
I enabled the "Web search" setting in the admin panel, selecting "google_pse" as the backend search engine to use. I created a Google Search API key, as well as a Google PSE Engine Id and entered them into Open-WebUI. I left the other settings at their defaults.
This did cause a "Web Search" button to show under my chat text input field. However, it seems to have no effect.
As a test, I tried asking, "What is the temperature in Tokyo today?" With web search button disabled, it responded with:
I currently don't have access to real-time weather data, so I can't provide the exact temperature in Tokyo for today. However, you can check the current temperature using a weather service like the **Japan Meteorological Agency (JMA)** website, **Weather.com**, or a weather app (e.g., Weather.com, AccuWeather, or the built-in weather app on your phone).
If you'd like, I can give you general information about Tokyo's typical temperatures for this time of year! Let me know.
Which is what I would expect. However, when I enabled the web search button, it says it searched 20 sites, and then responded with:
I currently don't have access to real-time weather data, so I can't provide the exact temperature in Tokyo for today. However, you can check the current temperature using a weather service like the **Japan Meteorological Agency (JMA)** website, **Weather.com**, or a weather app (e.g., Weather.com, AccuWeather, or the built-in weather app on your phone).
If you'd like, I can give you general information about Tokyo's typical temperatures for this time of year! Let me know.
So it's response was essentially identical, except it wasted a lot additional time searching the web. As the temperature in Tokyo is easily something you can find with a Google search, I can only assume the web results are not being forwarded to the LLM when it formulates its response. However, it's unclear what additional setting I need to change in order to fix this.
Steps to Reproduce
Go to /admin/settings
Click on Web Search
Toogle "Web Search" to "on".
Select "google_pse" as the search engine and enter your API key and PSE Engine ID.
Click Save.
Open a new chat.
Enable the "Web Search" button below the text input field.
Ask a question the LLM should only be able to find by searching the web.
Logs & Screenshots
Additional Information
No response
Originally created by @chrisspen on GitHub (May 2, 2025).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/13443
### Check Existing Issues
- [x] I have searched the existing issues and discussions.
- [x] I am using the latest version of Open WebUI.
### Installation Method
Pip Install
### Open WebUI Version
0.6.5
### Ollama Version (if applicable)
0.6.7
### Operating System
Ubuntu 24.04
### Browser (if applicable)
_No response_
### Confirmation
- [x] I have read and followed all instructions in `README.md`.
- [x] I am using the latest version of **both** Open WebUI and Ollama.
- [x] I have included the browser console logs.
- [x] I have included the Docker container logs.
- [x] I have listed steps to reproduce the bug in detail.
### Expected Behavior
Enabling "Web Search" should allow a model to search the web and use web results when formulating its response.
### Actual Behavior
I enabled the "Web search" setting in the admin panel, selecting "google_pse" as the backend search engine to use. I created a Google Search API key, as well as a Google PSE Engine Id and entered them into Open-WebUI. I left the other settings at their defaults.
This did cause a "Web Search" button to show under my chat text input field. However, it seems to have no effect.
As a test, I tried asking, "What is the temperature in Tokyo today?" With web search button disabled, it responded with:
I currently don't have access to real-time weather data, so I can't provide the exact temperature in Tokyo for today. However, you can check the current temperature using a weather service like the **Japan Meteorological Agency (JMA)** website, **Weather.com**, or a weather app (e.g., Weather.com, AccuWeather, or the built-in weather app on your phone).
If you'd like, I can give you general information about Tokyo's typical temperatures for this time of year! Let me know.
Which is what I would expect. However, when I enabled the web search button, it says it searched 20 sites, and then responded with:
I currently don't have access to real-time weather data, so I can't provide the exact temperature in Tokyo for today. However, you can check the current temperature using a weather service like the **Japan Meteorological Agency (JMA)** website, **Weather.com**, or a weather app (e.g., Weather.com, AccuWeather, or the built-in weather app on your phone).
If you'd like, I can give you general information about Tokyo's typical temperatures for this time of year! Let me know.
So it's response was essentially identical, except it wasted a lot additional time searching the web. As the temperature in Tokyo is easily something you can find with a Google search, I can only assume the web results are not being forwarded to the LLM when it formulates its response. However, it's unclear what additional setting I need to change in order to fix this.
### Steps to Reproduce
1. Go to /admin/settings
2. Click on Web Search
3. Toogle "Web Search" to "on".
4. Select "google_pse" as the search engine and enter your API key and PSE Engine ID.
5. Click Save.
6. Open a new chat.
7. Enable the "Web Search" button below the text input field.
8. Ask a question the LLM should only be able to find by searching the web.
### Logs & Screenshots
### Additional Information
_No response_
GiteaMirror
added the bug label 2026-04-19 22:43:53 -05:00
I see this quite a bit. Usually a regenerate will cause it to perform the search. Sometimes a 3rd time since it seems to only populate the RAG sometimes.
<!-- gh-comment-id:2848410961 -->
@geiseri commented on GitHub (May 3, 2025):
I see this quite a bit. Usually a regenerate will cause it to perform the search. Sometimes a 3rd time since it seems to only populate the RAG sometimes.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Originally created by @chrisspen on GitHub (May 2, 2025).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/13443
Check Existing Issues
Installation Method
Pip Install
Open WebUI Version
0.6.5
Ollama Version (if applicable)
0.6.7
Operating System
Ubuntu 24.04
Browser (if applicable)
No response
Confirmation
README.md.Expected Behavior
Enabling "Web Search" should allow a model to search the web and use web results when formulating its response.
Actual Behavior
I enabled the "Web search" setting in the admin panel, selecting "google_pse" as the backend search engine to use. I created a Google Search API key, as well as a Google PSE Engine Id and entered them into Open-WebUI. I left the other settings at their defaults.
This did cause a "Web Search" button to show under my chat text input field. However, it seems to have no effect.
As a test, I tried asking, "What is the temperature in Tokyo today?" With web search button disabled, it responded with:
Which is what I would expect. However, when I enabled the web search button, it says it searched 20 sites, and then responded with:
So it's response was essentially identical, except it wasted a lot additional time searching the web. As the temperature in Tokyo is easily something you can find with a Google search, I can only assume the web results are not being forwarded to the LLM when it formulates its response. However, it's unclear what additional setting I need to change in order to fix this.
Steps to Reproduce
Logs & Screenshots
Additional Information
No response
@geiseri commented on GitHub (May 3, 2025):
I see this quite a bit. Usually a regenerate will cause it to perform the search. Sometimes a 3rd time since it seems to only populate the RAG sometimes.