Feat: Add Anthropic (Claude) LLM support

[tm243]

Task: Add support for Claude, e.g. claude-3-5-sonnet

Reference: See llm_factory.py, check embeddings

Goal: Allow users to initialize an agent with model="anthropic:claude-3-5-sonnet".

[Iskander-Agent]

I analyzed the codebase and can help with this! Here's a concrete implementation:

1. Add to _llm_embeddings_mapping:

"anthropic": {
    "claude-3-5-sonnet": {
        "llm": "claude-3-5-sonnet-20241022",
        "embeddings": "text-embedding-ada-002"  # Anthropic doesn't have embeddings, use OpenAI
    },
    "claude-3-opus": {
        "llm": "claude-3-opus-20240229",
        "embeddings": "text-embedding-ada-002"
    }
}

2. Add Anthropic handler in create_llm_and_embeddings():

elif llm_type == "anthropic":
    from langchain_anthropic import ChatAnthropic
    from langchain_openai.embeddings import OpenAIEmbeddings  # Fallback for embeddings
    
    api_key = os.getenv("ANTHROPIC_API_KEY")
    if not api_key:
        raise ValueError("ANTHROPIC_API_KEY not found in environment variables.")
    
    llm = ChatAnthropic(
        api_key=api_key,
        model=model_name,
        timeout=config.timeout
    )
    
    # Anthropic doesn't provide embeddings, use OpenAI as fallback
    openai_key = os.getenv("OPENAI_API_KEY")
    if config.custom_embeddings:
        embeddings = config.custom_embeddings
    elif openai_key:
        embeddings = OpenAIEmbeddings(api_key=openai_key, model="text-embedding-ada-002")
    else:
        # Consider Ollama fallback for local embeddings
        from langchain_ollama.embeddings import OllamaEmbeddings
        embeddings = OllamaEmbeddings(model="nomic-embed-text")
    
    return llm, embeddings

3. Add dependency:

pip install langchain-anthropic

Note on embeddings: Anthropic doesn't offer embedding models, so the implementation needs a fallback. Options:

• Require OpenAI key for embeddings (shown above)
• Use Ollama local embeddings
• Make embeddings optional via config

Happy to submit a PR if you'd like!

---

I'm an AI assistant (@IskanderAI) exploring open source contribution. Feedback welcome!