I have searched for any existing and/or related issues.
I have searched for any existing and/or related discussions.
I have also searched in the CLOSED issues AND CLOSED discussions and found no related items (your issue might already be addressed on the development branch!).
I am using the latest version of Open WebUI.
Installation Method
Docker
Open WebUI Version
v0.8.8
Ollama Version (if applicable)
No response
Operating System
Linux
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 provided every relevant configuration, setting, and environment variable used in my setup.
I have clearly listed every relevant configuration, custom setting, environment variable, and command-line option that influences my setup (such as Docker Compose overrides, .env values, browser settings, authentication configurations, etc).
I have documented step-by-step reproduction instructions that are precise, sequential, and leave nothing to interpretation. My steps:
Start with the initial platform/version/OS and dependencies used,
Specify exact install/launch/configure commands,
List URLs visited, user input (incl. example values/emails/passwords if needed),
Describe all options and toggles enabled or changed,
Include any files or environmental changes,
Identify the expected and actual result at each stage,
Ensure any reasonably skilled user can follow and hit the same issue.
Expected Behavior
When the search_web tool is called, the search query used should be generated by the model defined in Admin Settings -> Interface -> Task Model
Actual Behavior
The Task Model is not called to generate the query, even though it is called to generate Titles/Tags/Followups (when enabled)
Using llama-proxy/llama.ccp as the backend model runners, configured in Configured in Admin Settings -> Connections as openapi endpoints
Model available in one of the backends chosen as both Local and External Task model:
Retrieval and Web Search Query Generation Enabled in UI (have also tried enabling via env variables), prompts left at default, but have tested with a customized Query Generation Prompt:
Title generation is functioning correctly -> Calls the listed model automatically and returns titles correctly. (Can see activity in llama and a OWUI log 2026-03-06 17:43:04.167 | INFO | uvicorn.protocols.http.httptools_impl:send:483 - 192.168.88.13:0 - "POST /api/v1/tasks/title/completions HTTP/1.1" 200)
When a model calls the search_web tool during a chat, the prompt used appears to be generated by the model itself (ie: small dumb models produce terrible queries, and large models produce good complete queries).
There is no activity seen on the selected Task Model and nothing in logs that indicates it was ever called.
I have tested with selecting various local models as the task model.
The models are set to Public and Local
Logs & Screenshots
Examples:
Additional Information
There are no Web Console logs shown during prompting of the model (the model calls the search_web tool during this)
OWUI Container logs when triggering the question as per screenshots:
(The model search query was "current month" and no request was made to the task model outside of generating a Title)
Originally created by @lbatalha on GitHub (Mar 6, 2026).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/22329
### Check Existing Issues
- [x] I have searched for any existing and/or related issues.
- [x] I have searched for any existing and/or related discussions.
- [x] I have also searched in the CLOSED issues AND CLOSED discussions and found no related items (your issue might already be addressed on the development branch!).
- [x] I am using the latest version of Open WebUI.
### Installation Method
Docker
### Open WebUI Version
v0.8.8
### Ollama Version (if applicable)
_No response_
### Operating System
Linux
### 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 **provided every relevant configuration, setting, and environment variable used in my setup.**
- [x] I have clearly **listed every relevant configuration, custom setting, environment variable, and command-line option that influences my setup** (such as Docker Compose overrides, .env values, browser settings, authentication configurations, etc).
- [x] I have documented **step-by-step reproduction instructions that are precise, sequential, and leave nothing to interpretation**. My steps:
- Start with the initial platform/version/OS and dependencies used,
- Specify exact install/launch/configure commands,
- List URLs visited, user input (incl. example values/emails/passwords if needed),
- Describe all options and toggles enabled or changed,
- Include any files or environmental changes,
- Identify the expected and actual result at each stage,
- Ensure any reasonably skilled user can follow and hit the same issue.
### Expected Behavior
When the `search_web` tool is called, the search query used should be generated by the model defined in Admin Settings -> Interface -> Task Model
### Actual Behavior
The Task Model is not called to generate the query, even though it is called to generate Titles/Tags/Followups (when enabled)
### Steps to Reproduce
Setup:
Docker Compose:
```
services:
openwebui:
image: ghcr.io/open-webui/open-webui:main-slim
container_name: openwebui
restart: unless-stopped
stop_grace_period: "120s"
environment:
- RAG_SYSTEM_CONTEXT=True
- CORS_ALLOW_ORIGIN=https://mydomain;http://127.0.0.1:33808
#- ENABLE_OTEL=True
#- ENABLE_OTEL_METRICS=True
- AIOHTTP_CLIENT_TIMEOUT=1600
networks:
- ml
ports:
- "127.0.0.1:33808:8080"
volumes:
- ./data:/app/backend/data
cpu_shares: 1024
mem_limit: '2G'
```
Using `llama-proxy`/`llama.ccp` as the backend model runners, configured in Configured in Admin Settings -> Connections as openapi endpoints
Model available in one of the backends chosen as both Local and External Task model:
<img width="1087" height="125" alt="Image" src="https://github.com/user-attachments/assets/8183857d-ab83-472b-a244-0b638acca9c4" />
Retrieval and Web Search Query Generation Enabled in UI (have also tried enabling via env variables), prompts left at default, but have tested with a customized `Query Generation Prompt`:
<img width="1649" height="684" alt="Image" src="https://github.com/user-attachments/assets/788986f6-bb46-4eb8-aa37-afb453340fce" />
Title generation is functioning correctly -> Calls the listed model automatically and returns titles correctly. (Can see activity in llama and a OWUI log `2026-03-06 17:43:04.167 | INFO | uvicorn.protocols.http.httptools_impl:send:483 - 192.168.88.13:0 - "POST /api/v1/tasks/title/completions HTTP/1.1" 200`)
When a model calls the `search_web` tool during a chat, the prompt used appears to be generated by the model itself (ie: small dumb models produce terrible queries, and large models produce good complete queries).
There is no activity seen on the selected Task Model and nothing in logs that indicates it was ever called.
I have tested with selecting various local models as the task model.
The models are set to `Public` and `Local`
### Logs & Screenshots
Examples:
<img width="1271" height="348" alt="Image" src="https://github.com/user-attachments/assets/afd1a2bb-a10a-462d-b9bc-b50ab5eb1787" />
<img width="1259" height="383" alt="Image" src="https://github.com/user-attachments/assets/20f48dfe-1579-4906-ad76-9c934164a9ed" />
<img width="1278" height="402" alt="Image" src="https://github.com/user-attachments/assets/703f1ea3-2476-446a-9f67-f537dfd252fa" />
### Additional Information
There are no Web Console logs shown during prompting of the model (the model calls the `search_web` tool during this)
OWUI Container logs when triggering the question as per screenshots:
```
2026-03-06 18:09:14.198 | INFO | uvicorn.protocols.http.httptools_impl:send:483 - 192.168.88.13:0 - "POST /api/v1/chats/8496cf2c-ed99-46ea-8e76-20efe0aa66a3 HTTP/1.1" 200
2026-03-06 18:09:14.248 | INFO | uvicorn.protocols.http.httptools_impl:send:483 - 192.168.88.13:0 - "GET /api/v1/chats/?page=1 HTTP/1.1" 200
2026-03-06 18:09:14.410 | INFO | uvicorn.protocols.http.httptools_impl:send:483 - 192.168.88.13:0 - "POST /api/chat/completions HTTP/1.1" 200
2026-03-06 18:09:14.466 | INFO | uvicorn.protocols.http.httptools_impl:send:483 - 192.168.88.13:0 - "GET /api/v1/chats/?page=1 HTTP/1.1" 200
2026-03-06 18:09:16.609 | INFO | ddgs.base:search:116 - response: https://en.wikipedia.org/w/api.php?action=opensearch&profile=fuzzy&limit=1&search=current%20month 200
2026-03-06 18:09:16.663 | INFO | ddgs.base:search:116 - response: https://grokipedia.com/api/typeahead?query=current+month&limit=1 200
2026-03-06 18:09:16.765 | INFO | ddgs.engines.wikipedia:extract_results:56 - response: https://en.wikipedia.org/w/api.php?action=query&format=json&prop=extracts&titles=Current%20Anthropology&explaintext=0&exintro=0&redirects=1 200
2026-03-06 18:09:16.942 | INFO | ddgs.base:search:116 - response: https://www.mojeek.com/search?q=current+month 403
2026-03-06 18:09:17.358 | INFO | ddgs.base:search:116 - response: https://search.yahoo.com/search;_ylt=JiLaIYuySv20Y7Ltkppbt3Os;_ylu=T03vkzoDN2vIVJCnxk_x4rIUm6_I03SlMcZBKfKcbQ6zd50?p=current+month 200
2026-03-06 18:09:18.391 | INFO | ddgs.base:search:116 - response: https://yandex.com/search/site/?text=current+month&web=1&searchid=4305220 200
```
(The model search query was "current month" and no request was made to the task model outside of generating a Title)
GiteaMirror
added the bug label 2026-04-25 09:27:50 -05:00
Expected Behavior
When the search_web tool is called, the search query used should be generated by the model defined in Admin Settings -> Interface -> Task Model
"Expected behaviour" here is inaccurate - when native tool calling is used, then the model in the chat calls the tool. By calling the tool, the model in the chat also has to provide the parameters for the tool (incl. the search query)
the task model should not do it in native tool calling mode. the task model is not used for tool calling in native tool calling mode.
<!-- gh-comment-id:4013245801 -->
@Classic298 commented on GitHub (Mar 6, 2026):
> Expected Behavior
> When the search_web tool is called, the search query used should be generated by the model defined in Admin Settings -> Interface -> Task Model
"Expected behaviour" here is inaccurate - when native tool calling is used, then the model in the chat calls the tool. By calling the tool, the model in the chat also has to provide the parameters for the tool (incl. the search query)
the task model should not do it in native tool calling mode. the task model is not used for tool calling in native tool calling mode.
Thank you for the clarification!
I did a lot of searching the docs and old issues/discussions, but it wasn't obvious to me that this was the expected behavior. Maybe adding a notice box to https://docs.openwebui.com/features/chat-conversations/web-search/agentic-search noting this would make it clear that it is intended?
<!-- gh-comment-id:4013479683 -->
@lbatalha commented on GitHub (Mar 6, 2026):
Thank you for the clarification!
I did a lot of searching the docs and old issues/discussions, but it wasn't obvious to me that this was the expected behavior. Maybe adding a notice box to https://docs.openwebui.com/features/chat-conversations/web-search/agentic-search noting this would make it clear that it is intended?
<!-- gh-comment-id:4013490431 -->
@Classic298 commented on GitHub (Mar 6, 2026):
more information can be found here: https://docs.openwebui.com/features/extensibility/plugin/tools/#tool-calling-modes-default-vs-native
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 @lbatalha on GitHub (Mar 6, 2026).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/22329
Check Existing Issues
Installation Method
Docker
Open WebUI Version
v0.8.8
Ollama Version (if applicable)
No response
Operating System
Linux
Browser (if applicable)
No response
Confirmation
README.md.Expected Behavior
When the
search_webtool is called, the search query used should be generated by the model defined in Admin Settings -> Interface -> Task ModelActual Behavior
The Task Model is not called to generate the query, even though it is called to generate Titles/Tags/Followups (when enabled)
Steps to Reproduce
Setup:
Docker Compose:
Using
llama-proxy/llama.ccpas the backend model runners, configured in Configured in Admin Settings -> Connections as openapi endpointsModel available in one of the backends chosen as both Local and External Task model:
Retrieval and Web Search Query Generation Enabled in UI (have also tried enabling via env variables), prompts left at default, but have tested with a customized
Query Generation Prompt:Title generation is functioning correctly -> Calls the listed model automatically and returns titles correctly. (Can see activity in llama and a OWUI log
2026-03-06 17:43:04.167 | INFO | uvicorn.protocols.http.httptools_impl:send:483 - 192.168.88.13:0 - "POST /api/v1/tasks/title/completions HTTP/1.1" 200)When a model calls the
search_webtool during a chat, the prompt used appears to be generated by the model itself (ie: small dumb models produce terrible queries, and large models produce good complete queries).There is no activity seen on the selected Task Model and nothing in logs that indicates it was ever called.
I have tested with selecting various local models as the task model.
The models are set to
PublicandLocalLogs & Screenshots
Examples:
Additional Information
There are no Web Console logs shown during prompting of the model (the model calls the
search_webtool during this)OWUI Container logs when triggering the question as per screenshots:
(The model search query was "current month" and no request was made to the task model outside of generating a Title)
@Classic298 commented on GitHub (Mar 6, 2026):
"Expected behaviour" here is inaccurate - when native tool calling is used, then the model in the chat calls the tool. By calling the tool, the model in the chat also has to provide the parameters for the tool (incl. the search query)
the task model should not do it in native tool calling mode. the task model is not used for tool calling in native tool calling mode.
@lbatalha commented on GitHub (Mar 6, 2026):
Thank you for the clarification!
I did a lot of searching the docs and old issues/discussions, but it wasn't obvious to me that this was the expected behavior. Maybe adding a notice box to https://docs.openwebui.com/features/chat-conversations/web-search/agentic-search noting this would make it clear that it is intended?
@Classic298 commented on GitHub (Mar 6, 2026):
more information can be found here: https://docs.openwebui.com/features/extensibility/plugin/tools/#tool-calling-modes-default-vs-native