Which openai app event allows model to run code?

Hi team,
I've tried a number of app events and model settings to try and process a prompt that requires OpenAI to run code.

Specifically:
Need it to include tools-enabled environment with Python/PyMuPDF/OpenCV-style processing, not just plain GPT text generation.

Please let me know, thanks!

Hi @Feather01,

Currently, Pabbly Connect does not offer any action event for OpenAI that allows direct code execution within the workflow. OpenAI steps in Pabbly Connect are limited to text (and structured JSON) generation and do not provide a tools-enabled execution environment such as Python, PyMuPDF, or OpenCV.

Even when using features like function calling (tools), the model can only suggest actions or return structured outputs—it does not execute code or process files internally within Pabbly Connect.

As a workaround, you can offload the code execution to an external service. You can send data from Pabbly Connect to an external application (such as a custom server, API, or code execution environment) via Webhook or API by Pabbly. That external system can run the required code and return the processed response back to Pabbly Connect.

To implement this, you can split the workflow into two parts:

1. In Workflow 1, use a Data Forwarder or API by Pabbly as the final step to send data to the external application where the code will be executed.
2. In Workflow 2, capture the response from the external application (via webhook trigger), and continue the workflow using the returned data as needed.

Oh I see. That's unfortunate.
Is this because of security concern and so its inclusion was intentionally left out, or is it something that hasn't been rolled out yet but the Pabbly team plan to introduce later on?

Want to get a gauge of the limitations - whether permanent or short term.

Thanks~

Hi @Feather01,

This limitation is primarily due to how the OpenAI API is designed.

Currently, the OpenAI API does not support direct code execution within the API request itself. The models can generate code or structured outputs, but they cannot run that code in a tools-enabled environment (such as Python, PyMuPDF, or OpenCV) as part of the API call.

What is possible:

• Use the model to generate code or instructions
• Then execute that code in an external environment (custom server, or code action)
• Return the processed result back to your workflow

So at the moment, this is not just a platform-specific restriction but a broader limitation of how the OpenAI API operates.

Please let us know if you need help setting up the external execution workflow.

Thanks for the insight.

Does Claude have the same API level limitation?

I see on Pabbly Connect there is an claude module but app events are very limited,not like OpenAI's.

Hi @Feather01,

Yes — Claude has the same limitation as OpenAI models at the API level.

Both can generate code or structured outputs, but cannot execute code (Python, PyMuPDF, OpenCV, etc.) within API calls. Tool/function calling only returns instructions—it doesn’t run them.

So even with Claude in Pabbly Connect, you’ll still need an external service (API/server) to handle actual code execution and send the result back.
This is an API-level constraint, not specific to Pabbly.

Got it