[WIP] OpenWebUI update

[przepeck]

🛠 Summary

CVS-183785
Changing models used in OpenWebUI, adding sections about new agentic features.

TODO: Update screenshots to use ovms-model model name instead of Godreign/llama-3.2-3b-instruct-openvino-int4-model

🧪 Checklist

• Unit tests added.
• The documentation updated.
• Change follows security best practices.
  ``

[Copilot]

Pull request overview

Updates the OpenWebUI integration demo documentation to use newer default models (including a VLM) and adds guidance for newer “agentic” OpenWebUI features like Web Search, Memory, and Code Interpreter.

Changes:

• Switched the primary chat model example to OpenVINO/gpt-oss-20b-int4-ov and standardized the OpenWebUI Model ID to ovms-model.
• Replaced the VLM example model with Junrui2021/Qwen3-VL-8B-Instruct-int4 and added a new screenshot for image upload.
• Added new documentation sections for Web Search, Memory/context, and Code Interpreter configuration in OpenWebUI.

Reviewed changes

Copilot reviewed 1 out of 13 changed files in this pull request and generated 11 comments.

File	Description
demos/integration_with_OpenWebUI/README.md	Updates model pull/config instructions and adds new OpenWebUI feature sections (Web Search, Memory, Code Interpreter).
demos/integration_with_OpenWebUI/upload_images.png	Adds/updates a screenshot used by the VLM “upload images” step.

[Copilot]

In the Docker VLM pull command, --model_repository_path is set to models (relative path inside the container), while the volume is mounted to /models. This likely writes to a non-persisted path and breaks the subsequent --add_to_config --config_path /models/config.json step. Use /models here for consistency with other Docker examples above.

Suggested change

	docker run --rm -u $(id -u):$(id -g) -v $PWD/models:/models openvino/model_server:weekly --pull --source_model Junrui2021/Qwen3-VL-8B-Instruct-int4 --model_repository_path models --model_name ovms-model-vl --task text_generation
	docker run --rm -u $(id -u):$(id -g) -v $PWD/models:/models openvino/model_server:weekly --pull --source_model Junrui2021/Qwen3-VL-8B-Instruct-int4 --model_repository_path /models --model_name ovms-model-vl --task text_generation

[Copilot]

Windows VLM section uses --pipeline_type VLM_CB, but the Linux/Docker command omits it. If AUTO doesn’t reliably detect VLM, the Docker path may start the wrong pipeline. Align the Linux command with Windows by specifying the same --pipeline_type (and keep both examples consistent).

Suggested change

	docker run --rm -u $(id -u):$(id -g) -v $PWD/models:/models openvino/model_server:weekly --pull --source_model Junrui2021/Qwen3-VL-8B-Instruct-int4 --model_repository_path models --model_name ovms-model-vl --task text_generation
	docker run --rm -u $(id -u):$(id -g) -v $PWD/models:/models openvino/model_server:weekly --pull --source_model Junrui2021/Qwen3-VL-8B-Instruct-int4 --model_repository_path models --model_name ovms-model-vl --task text_generation --pipeline_type VLM_CB

[Copilot]

Step 2 for VLM no longer tells the reader to select the ovms-model-vl model in OpenWebUI (it just says “Start a New Chat”). Without selecting the correct model, the subsequent image-upload chat steps may run against the default text model instead of the VLM endpoint.

[Copilot]

Typo: “blow” should be “below” in “blue Web Search icon blow”.

Suggested change

	1. Open new Chat (it should have blue Web Search icon blow)
	1. Open new Chat (it should have blue Web Search icon below)

[Copilot]

The “Reference” is currently written as an image link (![...](...)), which renders as a broken image instead of a hyperlink. Use a normal markdown link here.

Suggested change

	
	[Open WebUI Web Search documentation](https://docs.openwebui.com/features/chat-conversations/web-search/agentic-search/)

[Copilot]

Typos: “manageble” should be “manageable”.

Suggested change

	It's possible to have multiple manageble memory records.
	It's possible to have multiple manageable memory records.

[Copilot]

Typo: “Buildin Tools” should be “Built-in Tools” (or whatever wording matches the OpenWebUI UI label).

Suggested change

	3. In **Buildin Tools** section enable **Memory**
	3. In **Built-in Tools** section enable **Memory**

[Copilot]

Typo: “beggingng” should be “beginning” in the note about searching memory.

Suggested change

	> **Note**: There is no way to make searching memory default on the beggingng of the conversation in Open Web UI. User should tell model to use it to make it work.
	> **Note**: There is no way to make searching memory default on the beginning of the conversation in Open Web UI. User should tell model to use it to make it work.

[Copilot]

In the VLM Windows example, --pull uses --model_name ovms-model-vl, but --add_to_config points --model_path to Junrui2021/Qwen3-VL-8B-Instruct-int4. If pull mode follows the documented behavior (downloads into <model_repository_path>/<model_name>), the model path should match ovms-model-vl (or omit --model_name in the pull step). As written, --add_to_config will likely fail to find the pulled model directory.

[Copilot]

Similarly in the Docker VLM example, --pull sets --model_name ovms-model-vl, but --add_to_config uses --model_path Junrui2021/Qwen3-VL-8B-Instruct-int4. If pull stores the model under /models/ovms-model-vl, --model_path should point to ovms-model-vl (or the pull step should not override --model_name).