FR: Support for OpenAI-Compatible Endpoints in AI Assistant

[nichu42]

Dear PDF-XChange Editor Team,

I would like to request the addition of support for OpenAI-compatible endpoints in the AI Assistant feature. Currently, the Assistant only supports models from OpenAI, Anthropic, and Google, which limits users to a small set of proprietary, US-based providers.

Many users, especially in Europe, prioritize digital sovereignty, data privacy, and vendor neutrality.
Supporting OpenAI-compatible endpoints would allow us to use:
- Self-hosted or locally run LLMs
- European or region-specific LLM providers that adhere to GDPR and other local regulations
- Open-source models that align with organizational policies or ethical guidelines

Technical Note: OpenAI-compatible endpoints typically use the same API structure as OpenAI (e.g., `/v1/chat/completions`), so integration should be straightforward.

Thank you for considering this request. I believe it would significantly enhance the value of PDF-XChange Editor for a wide range of users, particularly those in regulated industries or with strict data governance requirements.

[Daniel - PDF-XChange]

Hello, nichu42

If i am not mistaken, so long as the API key follows the same structure as the base model, you should be able to utilize them in much the same way. I recall overhearing some testing of this with an "copilot" token, which operates through OpenAI as well. Please do try supplying your API key under the relevant section, and see if it works.

If you can cite specific models where this does not work, please do feel free to email [email protected] with the name(s) of the service/model. It will be passed along to the Dev team for review and hopefully implementation in the future.

Kind regards,

[nichu42]

I would need to define a different endpoint. The one from OpenAI seems to be hardcoded.

[Sean - PDF-XChange]

HI Nichu,

Okay, thank you - if we can assist you with this issue, or if you need further input from our team on it, please email [email protected] and we will proceed with helping you by email.

Kind regards,

[nichu42]

I don't need support. It's a feature request. I thought it's the correct place for this?

[Sean - PDF-XChange]

Hi Nichu,

I was just reiterating Dan's point above about citing specific models where what you mentioned does not work. Other than that, I don't think there is any way we can proceed here, beyond what has already been discussed?

Kind regards,

[nichu42]

I'll try again to explain what this request is about.

You support three providers at the moment:

image.png

You can currently choose OpenAI. But it only asks for an API key. The request will go to https://api.openai.com/v1, which is hardcoded into PDF-XChange.

I request to add fourth option: OpenAI compatible.
So you can use any provider with an OpenAI-compatible API. Which is basically everyone, including famous OpenRouter that provides access to hundreds of models. You would need to enter the endpoint (e.g. https://openrouter.ai/api/v1) manually + API key.

[Sean - PDF-XChange]

Hi Nichu,

Okay, thanks for the clarification - I'll ask the development team for their input here.

Kind regards,

[Alex - PDF-XChange]

Hi Nichu,
Thank you for showing that there is interest in this feature. It is definitely on our to-do list.

[And1]

I just want to say that this feature is essential. As the original poster stated, data ownership and privacy are essential reasons why many companies, including European but also American, seek out solutions like PDF-XChange (as opposed to Adobe who had the privacy nightmare PR a year or two ago). Please consider implementing this same reliability into your ai features by letting us specify the endpoint. It can be the same backend code as the openai api because, like the original poster (OP) said, it's OpenAI compatible as a popular standard nowadays, to access any other LLM endpoints/api's. This could be self-hosted on a LAN, or elsewhere, it gives us critical assurances to use for business purposes.

Thank you so much for your consideration of our requests, which will certainly last many many years as a favorable feature to anyone wanting private AI!

[Daniel - PDF-XChange]

Hello, And1

Thank you for adding your voice. This is indeed in our future plans, as noted above. I cannot promise a timeline, but it should come eventually. For now I can only ask for patience.

Kind regards,