AI-Powered GPT Connect

Note: The AI model is trained on data available up until a certain date. For AI to have the latest knowledge, enable web search.

Using AI-Powered GPT Connect

Prerequisites: Install the extension.

1. On the list of extensions, click Open next to AI-Powered GPT Connect.
2. On the Agents screen, click Create agent.
3. In the Details step:
   
   1. Enter a name for your agent. 
   2. Select a profile for this agent. See Profile Settings.
   3. Click Next.
4. In the Trigger step, select the data source which will be monitored, and the type of event that will trigger the agent. Click Next. 
5. In the Outgoing data items step:
   1. Enter your AI prompt. You can insert parameters that the extension should use when processing the prompt.
   2. Optional: Select one or more fields to be analyzed by AI.
   3. In the Evaluation right panel, click Evaluate. Once the evaluation result appears below, verify if it matches your expected outcome for this prompt. If not, revise the prompt and evaluate again until you are satisfied with the result.
   4. Click Next.
6. In the Incoming data items step, select the target fields where you want the AI response to be stored. The fields for mapping are based on the evaluation action from the previous step.
   Click Next.
7. Click Create. 

Profile Settings

In the profile configuration for the AI-Powered GPT Connect extension, the following settings are available.

OpenAI account

Choose the OpenAI account to be used by this extension:

• Built-in connection (no further setup) – the extension uses Caspio’s managed OpenAI API key for all requests. Credit consumption and billing follow Caspio’s standard Credits system. This option is selected by default for all accounts.
• Your own account – this option is available upon request. It allows you to connect the extension to your own OpenAI account using your personal API key. Select this option and provide your OpenAI API key in the new field that appears. When using your own key, Credits usage may differ from Caspio’s standard credit system.

AI model

When creating or editing a profile for the AI-Powered GPT Connect extension, you can select which OpenAI model the agent will use. Each model differs in reasoning depth, response quality, and Credits consumption.

• Recommended by Caspio – automatically uses the model that we recommend for optimal balance of accuracy, speed, and Credits consumption.
• GPT-5.2 – highest-performance model for the most demanding reasoning, analysis, and decision-making scenarios. Designed for maximum accuracy and depth, making it ideal for complex workflows and critical use cases. Consumes more Credits than GPT-5.1.
• GPT-5.1 – advanced reasoning model offering improved accuracy and consistency over GPT-5 mini, while remaining more cost-efficient than GPT-5.2. Well suited for complex tasks that require strong reasoning.
• GPT-5 – top-tier reasoning model for highly complex or multi-step tasks requiring deep understanding and detailed responses. Consumes significantly more Credits than GPT-4.1, GPT-4o, and lighter GPT-5 variants.
• GPT-5 mini – balanced model with stronger reasoning and higher output quality than GPT-4.1 mini, offering faster responses than GPT-5. Consumes significantly more Credits than GPT-4.1 mini.
• GPT-5 nano – lightweight model for simpler prompts and quick text generation, with better context handling than GPT-4.1 nano. Consumes significantly more Credits than GPT-4.1 nano, but remains the most cost-efficient option among GPT-5 models.
• GPT-4.1 – advanced model for high-accuracy processing, suitable for nuanced tasks involving long text or structured output.
• GPT-4.1 mini – balanced model for moderately complex tasks with faster response and lower consumption.
• GPT-4.1 nano – highly optimized for basic language tasks with minimal cost.
• GPT-4o – for processing complex requests that require high language precision and image interpretation.
• GPT-4o mini – optimized for fast performance and reduced costs on simpler requests.

For detailed information on GPT and OpenAI models, see OpenAI’s Models documentation.

Temperature

For all AI models prior to GPT-5.0, the Temperature setting controls how focused or creative the model’s responses are. Lower values produce more consistent and repeatable output, while higher values increase variability and creativity. Available options range from 0 — Strictly deterministic to 2.0 — Wild / chaotic. Higher temperatures may result in increased Credits consumption.

Top-P

For GPT-4.x and GPT-4o models, the Top-P setting controls how broadly the model considers possible responses when generating output. Lower values restrict responses to the most likely and predictable options, while higher values allow a wider range of expression and creativity. Available options range from 0.1 — Very deterministic to 1.0 — Maximum diversity. Higher Top-P values may result in increased Credits consumption.

Reasoning effort 

For GPT-5.0 models, the system displays the Reasoning effort setting instead of Temperature. This setting determines how much analytical depth the model applies when producing a response. The higher the reasoning effort, the more time AI will take to produce a response. Higher reasoning effort may result in increased Credits consumption.

Alongside standard reasoning effort degrees, the following options are available depending on the selected model: 

• None – available for GPT-5.1 and GPT-5.2. The AI generates direct responses without additional reasoning. Suitable for simple lookups, short answers, and cost-sensitive scenarios. 
• Minimal – available for GPT-5.0. The AI applies a very limited level of reasoning, balancing speed and cost with basic contextual understanding. 

Web search

Enable the Use web search for up-to-date responses option to allow GPT models to include latest online information in their responses. When enabled, the model can fetch relevant web content before generating an answer, improving accuracy for time-sensitive or frequently changing topics.

Web Search enhances prompts that depend on current data, such as news, prices, or live statistics. Responses that use Web Search may consume more Credits due to the additional processing involved.

Not available for GPT-4.1 nano model.

Knowledge

AI-Powered GPT Connect supports Knowledge at the Profile level, allowing app authors to define reusable AI agent context through Instructions and Files. Content defined in Knowledge is automatically included with every prompt sent to OpenAI for agents using the Profile, ensuring consistent behavior and responses aligned with the Profile’s intended purpose. 

The Instructions field defines how the AI should interpret prompts and respond across all requests associated with the Profile. Instructions are always applied and act as persistent guidance for the model. 

Instructions can be used on their own or together with uploaded files. When files are included, providing Instructions is strongly recommended so the model understands how the uploaded content should be interpreted and applied. Profiles may be saved with Instructions only, or with both Instructions and Files. 

The Files section allows you to upload documents that provide additional contextual information for the AI. Uploaded files are stored in OpenAI vector storage and indexed using File Search. 

Files are uploaded and indexing begins when the Profile is saved. Indexing is asynchronous and may take up to 24 hours to complete. During this time, newly uploaded files may not yet be available for retrieval in AI responses. 

Only text content is indexed and retrievable during prompt execution. Non-text elements such as images, diagrams, scanned pages, or decorative content are ignored. 

To ensure the best retrieval results, verify that: 

• Files contain plain, machine-readable text and are not scanned documents or image-based PDFs. 
• Content is structured with clear headings, paragraphs, and lists to improve semantic chunking and retrieval accuracy. 
• Documents are concise and topic-focused rather than large, multi-topic files. 
• Decorative elements such as screenshots, charts, or embedded images are removed, as they are not indexed. 
• PDF files contain selectable text and are not generated from OCR-only scans. 
• Files do not contain redundant or duplicate content. 
• Instructions clearly explain how the AI should interpret and prioritize the uploaded content when files are used.

Supported file types and limits:

• Supported file types: TXT, MD, DOCX, text-based PDF 
• Maximum file size: 10 MB per file 
• Maximum number of files: 100 per Profile 

Outgoing Data Items

• AI prompt: Create a prompt with table fields for the AI to process. Specify response length, value type, or if extra details should be excluded, and adjust for the destination field type (for example, max 255 characters).
• File: Include File or Attachment data type fields in the prompt to be analyzed by AI. The supported formats are: c, cpp, css, csv, doc, docx, gif, go, html, java, jpeg, jpg, js, json, md, pdf, php, pkl, png, pptx, py, rb, tar, tex, ts, txt, webp, xlsx, xml, zip.For png, jpeg, jpg, webp and non-animated gif files, the file size must not exceed 20 MB. For optimal image processing, the recommended resolution is 1024×1024 pixels or less. For other file types, the maximum allowed size is 512 MB.

In the Outgoing data items step, you can evaluate your AI prompt. The evaluation results offer the fields that can potentially be populated when this Agent is executed. Map the fields returned by the evaluation to your table fields in the next step.

Note: Evaluations use the Credits resource in your Caspio account.

Incoming Data Items

Fields available here depend on whether or not you performed the prompt evaluation in the previous step. The final output received from OpenAI depends on the data types of the target fields that you selected.

If you performed the AI prompt evaluation, all the fields that were returned in the evaluation are available for mapping. The table fields that can be selected depend on the data types of the fields returned in the evaluation.

If you did not perform the AI prompt evaluation, two fields are available: 

• Response – stores the result of the AI analysis 
• Response status – stores the success or failure status  

Resource Usage

This extension uses two types of resources:

• Integrations calls: An agent makes 3 integrations calls per task to send, receive, and log, with an additional call for each optional document of File data type (does not apply to Attachment data type).
• Credits: Caspio Credits are used for sending requests and receiving the AI response, as well as for using the prompt evaluation feature. The consumption depends on the prompt complexity, the size and complexity of the attached files, the selected model, the response size, and on whether or not web search is enabled.
  

When Knowledge files are uploaded, additional Credits usage applies. Uploaded Knowledge files consume Credits based on storage size, calculated at 50 credits per 1 GB per day, using the highest storage usage recorded during a day. Storage usage is prorated, with a minimum charge of 0.01 credit per day. In addition, when Knowledge files are used to answer a prompt, extra credits are consumed on top of the standard request-related usage.