This script wires an OpenAI chat model directly into a Zoho Creator form. The user types a prompt into a text field, Deluge calls the OpenAI API with invokeurl, and the generated text lands in an output field on the same record — no middleware, no external automation tool.

It handles three triggers: generate from a fresh prompt, refine an existing output with a follow-up instruction, and reset the form state.

Before you start

• An OpenAI API key and, if your account uses one, an organization ID.
• Three fields on the form: Text_Input (single line or multi line), Chat_Output (multi line), and Refine_Input (decision box).
• Store the key in a Creator variable or a connection rather than pasting it inline — the sample below references YOUR_API_KEY and YOUR_ORG_ID as placeholders.

Deluge script

// ============================================
// ZOHO CREATOR -> OPENAI CHAT COMPLETIONS
// Trigger: On user input of the Text_Input field
// ============================================

// --- Trigger 1: new prompt, no existing output -> generate ---
if(input.Text_Input != "" && input.Refine_Input == false && input.Chat_Output == "")
{
    headers = Map();
    headers.put("Content-Type","application/json");
    headers.put("Authorization","Bearer " + YOUR_API_KEY);
    headers.put("OpenAI-Organization", YOUR_ORG_ID);

    messages = List();
    userMsg = Map();
    userMsg.put("role","user");
    userMsg.put("content",input.Text_Input);
    messages.add(userMsg);

    data = Map();
    data.put("model","gpt-4o-mini");
    data.put("messages",messages);

    response = invokeurl
    [
        url    : "https://api.openai.com/v1/chat/completions"
        type   : POST
        parameters : data.toString()
        headers: headers
    ];

    choices = response.getJSON("choices");
    if(choices != null && choices.size() > 0)
    {
        message = choices.get(0).getJSON("message");
        input.Chat_Output = message.getJSON("content");
    }
    else
    {
        info "OpenAI error: " + response.toString();
    }
}
// --- Trigger 2: existing output + new instruction -> refine ---
else if(input.Text_Input != "" && input.Chat_Output != null && input.Chat_Output != "")
{
    headers2 = Map();
    headers2.put("Content-Type","application/json");
    headers2.put("Authorization","Bearer " + YOUR_API_KEY);
    headers2.put("OpenAI-Organization", YOUR_ORG_ID);

    sysMsg = Map();
    sysMsg.put("role","system");
    sysMsg.put("content","Rewrite the user's text according to their instruction. Return only the rewritten text.");

    priorMsg = Map();
    priorMsg.put("role","user");
    priorMsg.put("content","TEXT:\n" + input.Chat_Output + "\n\nINSTRUCTION:\n" + input.Text_Input);

    messages2 = List();
    messages2.add(sysMsg);
    messages2.add(priorMsg);

    data2 = Map();
    data2.put("model","gpt-4o-mini");
    data2.put("messages",messages2);

    response2 = invokeurl
    [
        url    : "https://api.openai.com/v1/chat/completions"
        type   : POST
        parameters : data2.toString()
        headers: headers2
    ];

    choices2 = response2.getJSON("choices");
    if(choices2 != null && choices2.size() > 0)
    {
        message2 = choices2.get(0).getJSON("message");
        input.Chat_Output = message2.getJSON("content");
    }
}

// --- Trigger 3: On user input of the Refine_Input decision field ---
if(input.Refine_Input == true)
{
    input.Text_Input = "";
    input.Refine_Input = false;
}

Notes

• Model name is a moving target. gpt-4o-mini is used here as a cost-effective default. Swap it for whichever model your account has access to; the request shape does not change.
• Guard the response. The original script indexed straight into choices. If the API returns an error object instead, that throws. The version above checks the array first and logs the raw response with info so you can see what came back.
• Rate limits and timeouts. Long prompts on a form-level trigger can exceed Creator’s script timeout. For heavy generation, move the call to a scheduled function or a standalone function invoked from a button.