AgentBridge

Appearance

LLM Providers ​

The Agent Bridge supports multiple AI providers for language model intelligence.

Supported Providers ​

1. Google Gemini (Default) ​

The primary reasoning engine, optimized for tool calling and agentic workflows.

  • Env Var: GOOGLE_API_KEY
  • Default Model: gemini-2.5-flash
  • Available Models:
    • gemini-2.5-flash - Fast, cost-effective (Recommended).
    • gemini-2.5-pro - Higher accuracy for complex reasoning.
    • gemini-1.5-pro - Previous generation.

2. OpenAI ​

Full support for OpenAI's GPT models.

  • Env Var: OPENAI_API_KEY
  • Default Model: gpt-4o
  • Available Models:
    • gpt-4o - Multimodal flagship model.
    • gpt-4-turbo - Previous generation.
    • gpt-3.5-turbo - Fastest, lowest cost.

3. AWS Bedrock (Disabled) ​

Currently disabled due to dependency conflicts with langchain-aws.

  • To enable, add langchain-aws to pyproject.toml and uncomment in llm_factory.py.

How Provider Selection Works ​

The agent selects a provider based on the model_provider parameter passed during invocation.

Code Path: server/app/services/agent/llm_factory.py

python
# The factory dispatches to provider-specific modules:
PROVIDER_MAP = {
    "gemini": get_gemini_llm,
    "openai": get_openai_llm,
    # "bedrock": get_bedrock_llm,  # Disabled
}

Adding a New Provider ​

  1. Create a new file: server/app/services/agent/providers/your_provider.py.
  2. Implement a get_your_provider_llm(model_name: str) function.
  3. Register it in PROVIDER_MAP in llm_factory.py.

Configuration ​

All providers are configured via environment variables in server/.env.

bash
# Example .env configuration
GOOGLE_API_KEY="AIzaSy..."
OPENAI_API_KEY="sk-..."