[Bug]: Fix `EmbeddingStrategy._get_embedding_llm_config_dict` ignoring local embeddings

[blghtr]

crawl4ai version

0.7.7

Expected Behavior

When EmbeddingStrategy is initialized without an explicit embedding_llm_config (or with None), I expect it to faithfully return None from _get_embedding_llm_config_dict(). This would allow the downstream get_text_embeddings utility to correctly switch to the local sentence-transformers implementation.

Current Behavior

The method _get_embedding_llm_config_dict contains a fallback mechanism that forces an OpenAI configuration if self.config is present but has no LLM config.

Relevant code in adaptive_crawler.py (lines 633-644):

    def _get_embedding_llm_config_dict(self) -> Dict:
        # ... check self.config ...
        
        # Fallback to default if no config provided
        return {
            'provider': 'openai/text-embedding-3-small',
            'api_token': os.getenv('OPENAI_API_KEY')
        }

Because this method always returns a dict, the logic in utils.py's get_text_embeddings never reaches the else block for local embeddings:

# utils.py
if llm_config is not None:
    # Uses LiteLLM (calls OpenAI)
else:
    # Uses sentence-transformers (UNREACHABLE via EmbeddingStrategy)

This effectively makes local embeddings dead code when using AdaptiveCrawler.

Is this reproducible?

Yes

Inputs Causing the Bug

- **Config**: `AdaptiveConfig(strategy="embedding", embedding_llm_config=None)` (intended for local mode).

Steps to Reproduce

1. Install `crawl4ai` with `sentence-transformers`.
2. Configure `AdaptiveCrawler` to use `"embedding"` strategy without providing `embedding_llm_config`.
3. Run `crawler.digest(...)`.
4. **Expected**: Uses local model.
5. **Actual**: Fails with OpenAI authentication error (or attempts to use OpenAI if key is in env), ignoring the user's intent to use local embeddings.

Code snippets

OS

Windows 11

Python version

3.11

Browser

No response

Browser version

No response

Error logs & Screenshots (if applicable)

No response

[Ahmed-Tawfik94]

Root Cause & Analysis

• In crawl4ai/adaptive_crawler.py, _get_embedding_llm_config_dict always returned a fallback OpenAI config when embedding_llm_config was None, so get_text_embeddings never saw None and always used remote embeddings. Local sentence-transformers path was unreachable.

Problematic code:

def _get_embedding_llm_config_dict(self) -> Dict:
    if hasattr(self, 'config') and self.config:
        config_dict = self.config._embedding_llm_config_dict
        if config_dict:
            return config_dict

    # Fallback to default if no config provided
    return {
        'provider': 'openai/text-embedding-3-small',
        'api_token': os.getenv('OPENAI_API_KEY')
    }

Suggestion

• Change the method to return None when no embedding config is provided (only return a dict for explicit configs), so get_text_embeddings can follow the local sentence-transformers branch. Add a regression test to assert _get_embedding_llm_config_dict() returns None when embedding_llm_config is None.