Feature: Enable programmatically pass in api_key besides reading from env

[aryasoni98]

This PR implements the ability to pass API keys programmatically to the EmbedText function, enabling portable AI agents that don't depend on host environment variables. This is particularly useful for CI/CD environments like Bitbucket pipelines where environment variables are managed through pipeline configuration.

Solution

Added api_key parameter to EmbedText function and extended the LLM configuration system to support programmatic API key passing across all LLM providers.

🔧 Core Functionality

• Added api_key parameter to EmbedText function spec
• Extended LLM configuration system with API key support for all providers
• Updated all LLM client implementations to accept API keys from config
• Maintained backward compatibility with environment variables

📁 Files Modified

Python Files

• python/cocoindex/functions.py - Added api_key parameter to EmbedText
• python/cocoindex/functions/_engine_builtin_specs.py - Added api_key parameter to EmbedText
• python/cocoindex/llm.py - Added new LLM config classes with API key support

Rust Files

• src/llm/mod.rs - Extended LlmApiConfig enum with new config types
• src/llm/voyage.rs - Updated Voyage client to accept API key from config
• src/llm/gemini.rs - Updated Gemini client to accept API key from config
• src/llm/anthropic.rs - Updated Anthropic client to accept API key from config
• src/llm/openai.rs - Updated OpenAI client to accept API key from config
• src/llm/litellm.rs - Updated LiteLLM client to accept API key from config
• src/llm/openrouter.rs - Updated OpenRouter client to accept API key from config
• src/llm/vllm.rs - Updated VLLM client to accept API key from config
• src/ops/functions/embed_text.rs - Updated EmbedText function to handle api_key parameter

🚀 Usage Examples

Before (Environment Variable Only)

# Required: export VOYAGE_API_KEY="your-api-key"
embed_text = cocoindex.functions.EmbedText(
    api_type=cocoindex.LlmApiType.VOYAGE,
    model="voyage-code-3"
)

After (Programmatic API Key)

# No environment variable required
embed_text = cocoindex.functions.EmbedText(
    api_type=cocoindex.LlmApiType.VOYAGE,
    model="voyage-code-3",
    api_key="your-api-key-here"
)

Bitbucket Pipeline Example

def create_codebase_indexing_pipe():
    """Create a codebase indexing pipeline for Bitbucket."""
    # Get API key from Bitbucket pipeline variables
    api_key = os.environ.get("VOYAGE_API_KEY")  # Set in Bitbucket pipeline
    
    # Create embedding function with programmatic API key
    embed_function = cocoindex.functions.EmbedText(
        api_type=cocoindex.LlmApiType.VOYAGE,
        model="voyage-code-3",
        api_key=api_key  # No need to set environment variable on host
    )
    
    return embed_function

Multiple API Types

# Voyage AI
voyage_embed = cocoindex.functions.EmbedText(
    api_type=cocoindex.LlmApiType.VOYAGE,
    model="voyage-code-3",
    api_key="voyage-api-key"
)

# OpenAI
openai_embed = cocoindex.functions.EmbedText(
    api_type=cocoindex.LlmApiType.OPENAI,
    model="text-embedding-3-small",
    api_key="openai-api-key"
)

# Gemini
gemini_embed = cocoindex.functions.EmbedText(
    api_type=cocoindex.LlmApiType.GEMINI,
    model="text-embedding-004",
    api_key="gemini-api-key"
)

This implementation maintains full backward compatibility:

• If no api_key is provided, the system falls back to environment variables
• Existing code continues to work without modification
• No breaking changes to the public API

🧪 Testing

• ✅ Verified Python syntax for all modified files
• ✅ Confirmed exact API signature from issue works
• ✅ Tested backward compatibility with environment variables
• ✅ Verified all LLM API types support programmatic API keys
• ✅ Confirmed Bitbucket pipeline use case is supported

📋 New LLM Config Classes

Added the following config classes with API key support:

• AnthropicConfig - For Anthropic Claude models
• GeminiConfig - For Google Gemini models
• VoyageConfig - For Voyage AI models
• LiteLlmConfig - For LiteLLM proxy models
• OpenRouterConfig - For OpenRouter models
• VllmConfig - For VLLM models
• Enhanced OpenAiConfig - Added api_key field

🔍 Implementation Details

Rust Side

• Extended LlmApiConfig enum to include all new config types
• Updated all LLM client constructors to accept api_config parameter
• Modified EmbedText function to create appropriate API configs based on api_key parameter
• Maintained fallback to environment variables when no config is provided

Python Side

• Added api_key parameter to EmbedText function spec
• Created new LLM config classes with API key support
• Updated LlmSpec to support all new config types
• Maintained existing API structure for backward compatibility

🎉 Result

Issue #994 is now 100% resolved.