Integrate OpenAI with ColdFusion: Native API and cfhttp Examples

OpenAI is usually the first integration ColdFusion teams try, and there are two ways to do it depending on your version.

Option A: Native ChatModel() (ColdFusion 2025 Update 8)

ChatModelConfig also exposes topP, topK, and seed for finer control, plus streaming support if you need the response to appear incrementally rather than all at once – useful for chat-style interfaces where users expect to see text appear as it’s generated.

chatModelConfig = {
    provider    : "openAI",
    modelName   : "gpt-5-nano",
    apiKey      : application.aiApiKey,
    temperature : 0.4,
    maxTokens   : 800,
    timeout     : 30
};

chatModel = ChatModel( chatModelConfig );
response  = chatModel.chat( "Write a 2-sentence product description for: " & productName );

writeOutput( encodeForHtml( response.message ) );

Option B: Direct API call with cfhttp (any other supported version, or Lucee)

httpService = new http();
httpService.setUrl( "https://api.openai.com/v1/chat/completions" );
httpService.setMethod( "post" );
httpService.addParam( type="header", name="Authorization", value="Bearer " & application.openAiApiKey );
httpService.addParam( type="header", name="Content-Type", value="application/json" );

requestBody = {
    "model"    : "gpt-4o-mini",
    "messages" : [
        { "role" : "user", "content" : "Write a 2-sentence product description for: " & productName }
    ],
    "temperature" : 0.4
};

httpService.addParam( type="body", value=SerializeJSON( requestBody ) );
result = httpService.send().getPrefix();

parsed = DeserializeJSON( result.fileContent );
writeOutput( encodeForHtml( parsed.choices[1].message.content ) );

Same outcome, more code to maintain – this is the approach we’ve used reliably on client projects for the past couple of years, well before the native API existed.

Function calling – letting GPT call your CFC methods 

Beyond plain chat, OpenAI’s models support function calling: you describe a function (name, parameters, description), and the model decides when to call it based on the conversation. In practice, this means you can expose a CFC method like getOrderStatus(orderId) and let the model call it directly when a user asks about their order – turning a Q&A bot into something that can actually retrieve real data mid-conversation. This requires more setup than a single chat call (defining the function schema, handling the model’s function-call response, executing your CFC method, then sending the result back to the model for a final answer) – worth a dedicated look once your basic integration is working.

Frequently Asked Questions

Which model name should I use? 

Check OpenAI’s current model list before hardcoding a name – model availability and naming change over time, and a model that’s current today may be deprecated in a year. 

Do I need to handle rate limits? 

Yes – wrap calls in retry logic with backoff for production use, the same way you’d handle any third-party API with rate limits. 

Is the native ChatModel() approach faster than cfhttp? 

Not meaningfully in terms of the actual API call latency – the benefit is less code to write and maintain, plus built-in streaming and structured output support. 

For help integrating OpenAI into your specific ColdFusion application, see our AI Integration Services with ColdFusion page.