issue: Model does not generate an answer after web search (shows raw <context> instead) #6790

New Issue

GiteaMirror · 2025-11-11T17:05:51-06:00

GiteaMirror commented

2025-11-11 17:05:51 -06:00

Originally created by @gamersalpha on GitHub (Oct 28, 2025).

Check Existing Issues

I have searched for any existing and/or related issues.
I have searched for any existing and/or related discussions.
I am using the latest version of Open WebUI.

Installation Method

Docker

Open WebUI Version

0.6.34

Ollama Version (if applicable)

0.12.6

Operating System

Debian 13

Browser (if applicable)

brave

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

The model should generate a normal answer using the retrieved web search context (RAG), for example: summarize the information found about the user question.
After the web search runs, I expect the assistant to reply in natural language.

Actual Behavior

The model does not generate an answer.
Instead, it only outputs the injected RAG context as raw XML-like text, for example:

<context>
<source id="1" name="Ariana Grande album">Ariana Grande Albums and Discography</source>
<source id="2" name="Ariana Grande album">Ariana Grande discography - Wikipedia</source>
</context>

<user_query>
Tell me what you can find about Ariana Grande’s latest album using web search.
</user_query>

No final answer is produced.

Steps to Reproduce

I am running Docker containers for:

open-webui (image: ghcr.io/open-webui/open-webui:main)

ollama (accessible at http://ollama:11434)

searxng (accessible at http://searxng:8080)
All services are on the same Docker network.

Environment variables for open-webui include:

ENABLE_RAG_WEB_SEARCH: True
RAG_WEB_SEARCH_ENGINE: "searxng"
SEARXNG_QUERY_URL: "http://searxng:8080/search?q=<query>&format=json"
ENABLE_OLLAMA: True
OLLAMA_API_BASE_URL: "http://ollama:11434"
DEFAULT_MODEL: "gemma3:270m"

In the UI, I select the model gemma3:270m from Ollama.

I ask a question that triggers web search, for example:

"Tell me what you can find about Ariana Grande’s latest album using web search."

Open WebUI successfully performs a web search through SearxNG.

I can see “Sources récupérées: 2”.

The logs show that documents were fetched, embedded, and added to a vector DB.

After that, instead of a summarized answer, the chat response from the model only prints the / / <user_query> block, and stops.

There is no actual generated answer.

Logs & Screenshots

Additional Information

Web search itself is working:
calling http://searxng:8080/search?q=test&format=json from inside the open-webui container returns valid JSON.

Retrieval/embedding also works: vectors are created and stored.

The only missing piece is generation: the model never turns the context into an answer.

Question:
Is this expected with small Ollama models (like gemma3:270m)?

Originally created by @gamersalpha on GitHub (Oct 28, 2025). ### 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 am using the latest version of Open WebUI. ### Installation Method Docker ### Open WebUI Version 0.6.34 ### Ollama Version (if applicable) 0.12.6 ### Operating System Debian 13 ### Browser (if applicable) brave ### 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 The model should generate a normal answer using the retrieved web search context (RAG), for example: summarize the information found about the user question. After the web search runs, I expect the assistant to reply in natural language. ### Actual Behavior The model does not generate an answer. Instead, it only outputs the injected RAG context as raw XML-like text, for example: ``` <context> <source id="1" name="Ariana Grande album">Ariana Grande Albums and Discography</source> <source id="2" name="Ariana Grande album">Ariana Grande discography - Wikipedia</source> </context> <user_query> Tell me what you can find about Ariana Grande’s latest album using web search. </user_query> ``` No final answer is produced. ### Steps to Reproduce I am running Docker containers for: open-webui (image: ghcr.io/open-webui/open-webui:main) ollama (accessible at http://ollama:11434) searxng (accessible at http://searxng:8080) All services are on the same Docker network. Environment variables for open-webui include: ``` ENABLE_RAG_WEB_SEARCH: True RAG_WEB_SEARCH_ENGINE: "searxng" SEARXNG_QUERY_URL: "http://searxng:8080/search?q=<query>&format=json" ENABLE_OLLAMA: True OLLAMA_API_BASE_URL: "http://ollama:11434" DEFAULT_MODEL: "gemma3:270m" ``` In the UI, I select the model gemma3:270m from Ollama. I ask a question that triggers web search, for example: "Tell me what you can find about Ariana Grande’s latest album using web search." Open WebUI successfully performs a web search through SearxNG. I can see “Sources récupérées: 2”. The logs show that documents were fetched, embedded, and added to a vector DB. After that, instead of a summarized answer, the chat response from the model only prints the <context> / <source> / <user_query> block, and stops. There is no actual generated answer. ### Logs & Screenshots <img width="1331" height="418" alt="Image" src="https://github.com/user-attachments/assets/941b9fe6-1b25-4bb4-a482-c8e8da2d0c35" /> ### Additional Information Web search itself is working: calling http://searxng:8080/search?q=test&format=json from inside the open-webui container returns valid JSON. Retrieval/embedding also works: vectors are created and stored. The only missing piece is generation: the model never turns the context into an answer. Question: Is this expected with small Ollama models (like gemma3:270m)?

GiteaMirror added the bug label 2025-11-11 17:05:51 -06:00

GiteaMirror closed this issue

2025-11-11 17:05:52 -06:00

GiteaMirror commented

2025-11-11 17:05:53 -06:00

@gamersalpha commented on GitHub (Oct 28, 2025):

OK i try with a model and works

@gamersalpha commented on GitHub (Oct 28, 2025): OK i try with a model and works <img width="1111" height="345" alt="Image" src="https://github.com/user-attachments/assets/7e4ca25c-e3ab-4511-b932-34c5ab9728fd" />

GiteaMirror referenced this issue

2026-04-19 20:49:57 -05:00

[GH-ISSUE #6790] Arena Model Needs a Confirmation Dialog #14488

GiteaMirror referenced this issue

2026-04-25 04:20:51 -05:00

[GH-ISSUE #6790] Arena Model Needs a Confirmation Dialog #30016

GiteaMirror referenced this issue

2026-05-05 14:22:15 -05:00

[GH-ISSUE #6790] Arena Model Needs a Confirmation Dialog #53154

GiteaMirror referenced this issue

2026-05-13 01:30:55 -05:00

[GH-ISSUE #6790] Arena Model Needs a Confirmation Dialog #69092