OpenAI Piece: Assistants, DALLE-3, and more!

Hello everyone, I’m MoShizzle, part of the pieces development team at Activepieces. We’re releasing a big update to our OpenAI piece! OpenAI released a few more tools to play with, and we were excited to get them in your hands.

:robot: Use the power of assistants

You can now use assistants and their many tools using the Ask Assistant action

:camera_flash: Stunning visuals and images in your flows

Use OpenAI’s latest text-to-image model, DALLE-3 to generate amazing imagery for your flows.

:speech_balloon: Generate audio from text with ease

Enter the text you want to hear, pick your preferred voice and let the Text-to-Speech action turn it into voice.

:memo: Keep your conversation history with ChatGPT

Want ChatGPT or your Assistant to remember what you asked it? Just fill the memory key with any random string of your choosing! (Using the same key in two different steps gives them the same memory)

  1. Enter memory key

  2. Ask GPT something

  3. Ask it to summarize your conversation

:eye_in_speech_bubble: Ask ChatGPT about an image

Using the Vision Prompt action, you can ask ChatGPT to tell you something about an image!

Let us know if there’s more you’d like to see from our OpenAI piece.

Till next time.

5 Likes

Look great!

Could this explain why I am receiving this error all of a sudden, Invalid model: gpt-4-1106-preview,

Even though it worked earlier today (the last successful run was 17:08 GMT), it still works in the OpenAI playground but not in ActivePieces.
playground

Hi Gunner,

It seems that the version we’re using for the OpenAI package is causing this issue, let me look into it for a potential fix.

2 Likes

Awesome, I appreciate that. I have a number of flows all set up to use it at present, so it would save me having to go and edit them all. :slight_smile:

1 Like

MoShizzle Bringing the Fire! Can’t wait to test it out :slight_smile:

2 Likes

Hi @MoShizzle,

Did you have any luck finding the issue?

Or will I have to use a different model for the time being? It will mean adjusting many flows, but I can’t afford to have them not running.

Kind regards,
GunnerJnr

Hello Gunner,

Sorry for the delay. The fix is released. Try it out and let me know if it’s solved for you (make sure your OpenAI piece is v0.3.13)

1 Like

Thank you. It is working again now.

1 Like

Love this! Very good job! I’ll love the assistants and the memory. I hadn’t noticed before I read this thread.

1 Like

Hi @MoShizzle

Since the fix has been implemented, I wanted to alert you to another minor issue that has arisen. (There is no need to panic.)

Issue Summary: When working with a workflow that involves, for example, ‘Google Sheets’, a specific problem arises with the ‘New Row’ trigger and switching between different workflow pieces. The issue pertains to the persistence of the ‘Tab’ state within a selected Google Sheet.

Detailed Description: In a workflow where one piece involves triggering an action based on a ‘New Row’ in a Google Sheet and another piece involves an action like ‘OpenAI - Ask ChatGPT’, there’s an observed inconsistency. If, for any reason, there is a need to switch back to the Google Sheets piece (for modifications or other reasons), the previously selected ‘Tab’ from the Google Sheet is not retained. It requires reselection upon every return to this piece. Notably, while the specific ‘Tab’ selection is lost, the workflow does seem to consistently remember the selected sheet itself.

Impact: This issue significantly hinders the efficiency and usability of the workflow. Every time users navigate back to the Google Sheets component, they must reselect the appropriate tab, disrupting the workflow continuity. This inconvenience might seem minor at first glance, but it has substantial implications, especially in complex workflows. For instance, if a subsequent step involves ‘Google Sheets - Update Row’, the failure to remember the selected sheet tab necessitates reselecting it each time, leading to additional manual input. This means in scenarios where multiple outputs (like 20 different ChatGPT responses) need to be entered into the sheet; users must repetitively re-enter all these details upon each navigation between workflow pieces. Such repetitive tasks not only slow down the process considerably but also increase the likelihood of errors, thereby affecting the overall efficiency and accuracy of the workflow.

Kind regards,
GunnerJnr

1 Like

Hi @GunnerJnr, thanks for reporting the issue!

This seems to be an issue with Google Sheets, not OpenAI, as multiple users have reported it in their flows as well - we are aware of it and a fix will be released tomorrow.

Sorry for the inconvenience!

Hi @MoShizzle,

Ok, thanks for the update. I only noticed it happening for me personally since the fix. Before that, it always remembered state, for me personally at least. But in any case, as long as you are aware, we know it’s in capable hands.

Thanks.

1 Like

Hi @MoShizzle,

I have encountered a depreciation issue today when wanting to use the vision api module:
“404 The model gpt-4-vision-preview has been deprecated, learn more here: https://platform.openai.com/docs/deprecations

the gpt 4 vision problem is still not fixed :frowning: