Configuration (config.yaml)

Relevant source files

The following files were used as context for generating this wiki page:

• config.yaml
• main.py
• agents.py
• searcher/sementic_search.py
• README.md

The config.yaml file serves as the central configuration hub for the CoI-Agent system. It defines essential parameters for connecting to Large Language Model (LLM) providers (OpenAI or Azure OpenAI), setting up embedding models, and authorizing access to external research APIs like Semantic Scholar.

This configuration is critical for the initialization of the DeepResearchAgent and ReviewAgent, as well as the SementicSearcher module, which rely on these settings to perform automated literature reviews and idea generation.

Configuration Loading and Environment Integration

At runtime, the application reads config.yaml and maps its values to system environment variables. This allows various modules to access secrets and endpoints globally via standard environment lookups.

Initialization Flow

The application initializes by loading the YAML file and updating os.environ. If a value in the configuration file is an empty string, it is skipped to prevent overwriting existing system variables.

The flow ensures that environment variables are populated before the get_llms() function is called to instantiate the model clients.
Sources: main.py:12-20, agents.py:17-27

LLM Provider Settings

The system supports two primary LLM provider paths: Azure OpenAI and standard OpenAI. This is toggled via the is_azure boolean flag.

Option	Description	Relevant Module
is_azure	Boolean flag (True/False) to switch between Azure and OpenAI.	main.py, agents.py
AZURE_OPENAI_ENDPOINT	The base URL for the Azure OpenAI service.	openai_llm client
AZURE_OPENAI_KEY	Authentication key for Azure services.	openai_llm client
AZURE_OPENAI_API_VERSION	Specific API version for Azure requests.	openai_llm client
OPENAI_API_KEY	Authentication key for standard OpenAI API.	openai_llm client
OPENAI_BASE_URL	Custom base URL for OpenAI (useful for proxies).	openai_llm client

Sources: config.yaml:3-13, README.md:50-60

Model Selection and Roles

The project distinguishes between "Main" and "Cheap" models to balance performance and cost during complex research tasks.

Tiered Model Configuration

The configuration defines two specific tiers:

• MAIN_LLM_MODEL: Used for high-reasoning tasks like judging ideas, refining experiments, and generating final research directions.
• CHEAP_LLM_MODEL: Used for token-heavy tasks like summarizing paper content, extracting entities, and providing initial reviews.

External Service API Keys

To function as a research agent, the system requires access to the Semantic Scholar Academic Graph.

• SEMENTIC_SEARCH_API_KEY: Required to authenticate requests to the Semantic Scholar API. This key is retrieved by the SementicSearcher class to perform asynchronous paper lookups and metadata retrieval.
• EMBEDDING_API_SETTINGS: Separate endpoints and keys can be provided for embedding models (e.g., for calculating cosine similarity during reranking). If left empty, the system defaults to using the main LLM configuration.

Searcher Configuration Table

Parameter	Function	Default/Requirement
SEMENTIC_SEARCH_API_KEY	Authorizes API calls to api.semanticscholar.org	Mandatory for search
EMBEDDING_MODEL	Specific model used for vector representations	Optional (Defaults to Main)
EMBEDDING_API_ENDPOINT	Custom endpoint for embedding generation	Optional

Sources: config.yaml:1-20, searcher/sementic_search.py:84-86, README.md:47-49

Conclusion

The config.yaml file is the foundational configuration layer of the CoI-Agent. By centralizing LLM provider details, model tiers, and external API keys, it allows the DeepResearchAgent and ReviewAgent to dynamically adapt to different computing environments and cost constraints while maintaining access to the scientific literature required for the Chain-of-Ideas process.