This add-on lets you create Ollama environments so you can host your own models or use a distant server running an Ollama instance.

Once the add-on is installed, you have nothing more to do on the add-on side. Go back to AI Engine, and in the Environments for AI settings, you will now be able to select the type “Ollama“, then add your endpoint and refresh the models list.

You can have your Ollama instance running on the same server as your WordPress server, so you will have to enter your server’s local address as the endpoint.

Function Calling

The “Functions” section in your chatbot settings will only appear if the model you have selected is noted as compatible with function calling.

Recent versions of Ollama say what each model can do (tools, vision, embedding), and the add-on uses that directly when you click Refresh Models. Older Ollama servers don’t send this information, so the add-on falls back on a short list of known models (llama3.1, llama3.2, mistral-nemo and a few others). If a model is not detected, you can force the function-calling tag yourself with a filter. There is also mwai_ollama_vision_models, which works the same way for vision.

add_filter( 'mwai_ollama_functions_models', 'my_custom_ollama_functions_models', 10, 3 );

/**
 * Adds custom models to the list of Ollama models that support functions.
 *
 * @param array $models The current list of model prefixes that support functions.
 * @param array $model_info The model information from the Ollama /api/tags endpoint.
 * @param array $details The detailed model information from the Ollama /api/show endpoint.
 * @return array The modified list of model prefixes.
 */
function my_custom_ollama_functions_models( $models, $model_info, $details ) {
    // Add 'deepseek-coder' and 'qwen2.5' to the list of models supporting functions
    $models[] = 'deepseek-coder';
    $models[] = 'qwen2.5';
    
    return $models;
}

These filters are made to return all model prefixes that should be function-calling compatible. You can either set the whole model name or just the beginning of it; for example, if you add “deepseek-” all the models whose name starts with “deepseek-” will get the function tag.