[BUG] WebsiteSearchTool always uses OpenAI despite Ollama being specified (both llm and embedder)

[MB-Finski]

Description

Following the docs here: https://docs.crewai.com/en/tools/search-research/websitesearchtool#customization-options

However, no matter what is set as the embedder and llm, WebSiteSearchTool always requires an OpenAI api key to be defined.

Steps to Reproduce

1. Install crewai["tools"]
2. Run the code provided in the docs
3. The following error appears: "Value error, The OPENAI_API_KEY environment variable is not set."

Specifying a dummy key leads to a failing request to openai.

Expected behavior

Allow setting Ollama as the llm and embeddings provider.

Screenshots/Code snippets

website_tool = WebsiteSearchTool(
    config=dict(
        llm=dict(
            provider="ollama", # or google, openai, anthropic, llama2, ...
            config=dict(
                model=main_model,
                base_url=ollama_api_base_url,
            ),
        ),
        embedder=dict(
            provider="ollama", # or openai, ollama, ...
            config=dict(
                model_name="embeddinggemma:latest",
                base_url=ollama_api_base_url,
                api_key="ollama",
            ),
        ),
    )
)

Operating System

Other (specify in additional context)

Python Version

3.12

crewAI Version

1.6.1

crewAI Tools Version

1.6.1

Virtual Environment

Venv

Evidence

pydantic_core._pydantic_core.ValidationError: 1 validation error for WebsiteSearchTool
  Value error, The OPENAI_API_KEY environment variable is not set. [type=value_error, input_value={'config': {'llm': {'prov...'RETRIEVAL_DOCUMENT'}}}}, input_type=dict]
    For further information visit https://errors.pydantic.dev/2.12/v/value_error

Possible Solution

None

Additional context

OS: Debian 13

[MB-Finski]

With regards to the AI generated PR: FWIW, I also tried passing in embedding_model but that fails, too.

[publicmutiny]

With regards to the AI generated PR: FWIW, I also tried passing in embedding_model but that fails, too.

I passed in embedding_model and it successfully switched to recognise Ollama as the embedder.

Code

from crewai_tools import WebsiteSearchTool

tool = WebsiteSearchTool(
    website='https://github.com',
    config=dict(
        llm=dict(
            provider="ollama",
            config=dict(model="llama2"),
        ),
        embedding_model=dict(
            provider="ollama",
            config=dict(model_name="nomic-embed-text")   # or another embedding model
        )
    )
)

Explanation:
rag_tool.py never checks the config for an 'embedder' dict, it only checks for 'embedding_model' (lines 138-145).

Fix:
Either the docs need updating to use embedding_model rather than embedder, or lib/crewai-tools/src/crewai_tools/tools/rag/rag_tool.py](url) needs updating to accept embedding_model instead of embedder.

[MB-Finski]

Still can't get it to work with embedding_model. This is likely a config issue on my side. When using the tool with an agent I get the following error:

pydantic_core._pydantic_core.ValidationError: 1 validation error for WebsiteSearchTool
  Value error, An embedding function already exists in the collection configuration, and a new one is provided. If this is intentional, please embed documents separately. Embedding function conflict: new: ollama vs persisted: openai [type=value_error, input_value={'collection_name': 'rag_...logservice_port=50052))}, input_type=dict]
    For further information visit https://errors.pydantic.dev/2.12/v/value_error

This is even though I haven't specified either an embedder or an embedding model anywhere else.

[greysonlalonde]

I'll update the docs and add backward compatibility here.

Value error, An embedding function already exists in the collection configuration... is a chromadb-related exception that pops up when you switch between providers in a shared collection. Try deleting that collection to get unblocked - we'll have a approach to this in a future version.