Configuring Connectors

Connectors are a component in the AVISE framework that contain the logic for making requests and receiving responses from AI models. When using AVISE, you need to pass a Connector configuration JSON file via the CLI argument --connectorconf.

AVISE comes with multiple prebuilt configuration files for commonly used AI model hosting services:

Ollama Connector

Ollama is a widely used software for running language models. To use AVISE with Ollama models, you can modify the existing avise/configs/connector/languagemodel/ollama.json configuration file:

avise/configs/connector/languagemodel/ollama.json

{
    "target_model": {
        "connector": "ollama-lm",
        "type": "language_model",
        "name": "phi3:latest", //ADD NAME OF THE OLLAMA MODEL TO TEST HERE
        "api_url": "http://localhost:11434", //Ollama default
        "api_key": null
    }
}

In the Connector configuraiton files, target_model defines the configurations for the model you wish to test and evaluate. target_model requires the following subfields:

• "connector": Name of the connector to use (See available connectors by running CLI command avise --connector_list)

• "type": Type of the AI model (e.g. language_model, continual_learning, multi_modal)

• "name": Name of the model (For Ollama connector, use the ollama model name)

• "api_url": URL for the API endpoint (http://localhost:11434 is the default when running Ollama models locally)

• "api_key": Optional API authorization key if required by the API Endpoint. Leave as ``null`` if not needed

OpenAI Connector

To use AVISE with OpenAI models, you can modify the existing avise/configs/connector/languagemodel/openai.json configuration file:

avise/configs/connector/languagemodel/openai.json

 {
     "target_model": {
         "connector": "openai-lm",
         "type": "language_model",
         "name": "gpt-4o-mini",
         "api_url": null,
         "headers": null,
         "api_key": "YOUR_OPENAI_API_KEY_HERE"
     }
 }

In the Connector configuration files, target_model defines the configurations for the model you wish to test and evaluate. target_model requires the following subfields:

• "connector": Name of the connector to use (See available connectors by running CLI command avise --connector_list)

• "type": Type of the AI model (e.g. language_model, continual_learning, multi_modal)

• "name": Name of the model (For OpenAI connector, use the openai model name)

• "api_url": URL for the API endpoint

• "headers": Optional Headers for the API request.

• "api_key": OpenAI Authentication API Key.

Generic REST API Connector

With the Generic REST API Connector, you can connect AVISE to any RESTful API Endpoint. To configure the Generic REST API Connector, modify the existing avise/configs/connector/languagemodel/genericrest.json configuration file:

avise/configs/connector/languagemodel/genericrest.json

 {
     "target_model": {
         "connector": "generic-rest-lm",
         "type": "language_model",
         "name": "<PLACEHOLDER_NAME>",
         "api_url": "PLACEHOLDER_URL",
         "headers": null,
         "api_key": null,
         "response_field": "KEY_OF_GENERATED_RESPONSE",
         "method": "POST"
     }
 }

In the Connector configuraiton files, target_model defines the configurations for the model you wish to test and evaluate. target_model accepts the following subfields:

• "connector": Name of the connector to use (See available connectors by running CLI command avise --connector_list)

• "type": Type of the AI model (e.g. language_model, continual_learning, multi_modal)

• "name": Name of the model. (Optional for Generic REST Connector)

• "api_url": URL for the API endpoint

• "headers": Optional Headers for the API request.

• "api_key": OpenAI Authentication API Key.

• "response_field": The field in which the response of the AI model is contained in the API request response. E.g. if the API response is:

  {status: "success", message: "example message", data: {"model_name": "Gemini 2.5", "prompt": "Hi, what is your name?", "response": "Hello there! I'm Gemini, developed by Google."}}, then the response_field would be ["data"]["response"].

• "method": API Method to use.