I’m encountering an issue in my automation workflow with Pabbly where the OpenAI Assistant is returning a hyperlink instead of a direct downloadable URL in plain text. This is causing my workflow to fail, as Pabbly cannot handle hyperlinks in this context and requires plain text URLs for proper functionality.
Here’s what I’ve tried so far:
Reviewed the API response formatting to ensure it should return plain text URLs.
Consulted with the Pabbly support team, who confirmed that the issue originates from the response provided by the OpenAI Assistant.
Explored workarounds, but none have proven effective for my specific use case.
Has anyone faced a similar issue or found a reliable solution for ensuring the OpenAI Assistant outputs a direct downloadable URL in plain text instead of a hyperlink? Any insights, suggestions, or guidance would be greatly appreciated.
When using OpenAI, you cannot retrieve the image or file download URL directly. OpenAI’s team has confirmed they are sending the URL, but during several test in assistant playground, I am receiving a hyperlink instead of a direct link.
Even after adjusting the prompt to request a raw link, the response still provided a hyperlink.
If OpenAI can provide a direct downloadable link instead of a hyperlink, it should work correctly in Pabbly workflow as well.
OpenAI does not provide assistants output files in an end-user-downloadable format.
Files, such as generations from code interpreter, are a file_id created in your organization storage, with purpose ‘assistants_output’. (They should be project-scoped, but the idea of scoping and limiting access to user data seems to be half-baked and abandoned.)
The assistants playground web UI re-hosts and rewrites AI output like your application must. This is undesirable and confusing. An API developer would want an API call response inspector playground, not an obfuscated “dummies” consumer app.
You can even instruct a citation-like generation with nothing behind it to be generated, and the web UI will rewrite it.
You must disregard the assistants “playground” for developing solutions. Here is how to get output files produced by an assistant.
You would want to upload to your own temporary web storage from OpenAI’s storage in order to rewrite a response with a user-accessible URL, or have an event link that retrieves from the metadata provided and your database.
I am creating a workflow with an AI agent that performs several actions using the openai assistant creation functionality.
The main problem/advantage is that there is no storage involved when my agent is trying to get the file, but everything is managed on the fly to the destination.
The playground environment is perfect for creating and customising such actions and it would be great and so much easier if it could just print the temporary url it generates in the playground as for my specific case I don’t need long term storage but just the time to grab the file and send it over the next step.
After a lot of debugging it looks like I finally managed to achieve this despite the current limitation of the assistant playground, with a python script that fetches the binary from OpenAI and uploads it directly the next destination.
Thank you for giving me the final push of inspiration and useful resources.