Does AI Positron support DeepSeek?

[liyan]

Hi I know AI Positron is based on chatgpt. Is there a possibility that you are also developing it based on DeepSeek? I am a user in China and has limited access to chatgpt. It would be nice if there is an ai assistant based on deepseek.
Thank you!!!!

[Radu]

Hello,

The AI Positron Enterprise add-on allows you to configure a connector based on OpenAI.
https://www.oxygenxml.com/doc/ug-addons ... prise.html
As Deepseek's API is 100% compatible with OpenAI it should work to connect the add-on to the Deepseek LLM using the OpenAI connector.
I tried connecting AI Positron to the "deepseek-chat" model not so long ago. Deepseek does not have support for vision (for reading image content) so some of our actions which require this will not work. Also I could not manage to make the support for function calls work so some complex actions that we have predefined will also not work. But most of our predefined AI actions should work.

Regards,
Radu

[liyan]

Thank you for your reply!!! I got a new message from the IT of my company: we have an internal fine-tuned deepseek model and that model uses Beartoken authentication method. If I want to connect the AI Positron to our internal deepseek model using the openai connector, do I need to modify the source code of your plugin? Do you allow this modification or do you offer an option to modify?

[liyan]

Hi Radu,
I learned that you tried connecting AI Positron to the "deepseek-chat" model not so long ago. Could you share the procedure for implementing this? Thank you very much!!

[Radu]

Hi,
So assuming that from the Oxygen main menu Help->"Install new add-ons" you install the "AI Positron Assistant Enterprise" add-on and get a trial license for it:
https://www.oxygenxml.com/ai_positron/r ... enterprise

Then if you start Oxygen, in the Oxygen Preferences->"Plugins / Oxygen AI Positron Assistant Enterprise / AI Service Configuration" I had to configure the Deepseek connection (directly to the Deepseek public AI) something like:

deepseek.png

Regards,
Radu

[liyan]

Hi Radu
Thank you for your reply.
As deepseek does not provide a public API key, my company deployed a deepseek model in our internal network. The model uses a beartoken authentication method. In this case, to connect AI Positron to deepseek using openai connector, the IT said that we may need to modify the source code of the plugin. Does your company offer an option to do that? How do I implement this if the model uses a beartoken authentication method?

[Radu]

Hi,
Whatever value you set in the API Key field when defining the OpenAI connection is passed as an HTTP request property ""Authorization"=>"Bearer " + AI_KEY_VALUE". So this is how it works, with a bearer token, with the HTTP authorization header being passed the value of the API key that you set in the preferences.
If that is not suitable for you, maybe you can show me how a small example of using "curl" to connect to your AI engine would look like.
Regards,
Radu

[liyan]

Hi Radu
Thank you for your reply.
I don't know if I understand it correctly. The AI Positron automatically coverts the value you enter in the API key field into a Bearer Token format when configuring the openAI connection. Is this right?
By the way, I do not know how to use curl.
Could you answer me another question? The IT in my company explained that in the future, they will switch the authentication method from bearertoken to basic. In this case, how do this work? Same as bearertoken? The AI Positron will also automatically converts the value entered in the API key field?

Sincerely
liyan

[Radu]

Hi Ilyan,
So:

I don't know if I understand it correctly. The AI Positron automatically coverts the value you enter in the API key field into a Bearer Token format when configuring the openAI connection. Is this right?

Correct

By the way, I do not know how to use curl.

Curl is a command line utility for sending HTTP requests.
I was hoping your IT colleagues could give you/us a sample about how using curl one can access your AI server. In this way it would be more clear to me how the AI Positron add-on should behave.

Could you answer me another question? The IT in my company explained that in the future, they will switch the authentication method from bearertoken to basic. In this case, how do this work? Same as bearertoken? The AI Positron will also automatically converts the value entered in the API key field?

Basic authentication is a different mechanism. I would envision this to work by you no longer providing an AI Key in the preferences, then when you invoke an action you would receive an authentication dialog from Oxygen asking for username/password credentials. Then Oxygen would cache your user/password credentials and reuse them without asking you every time you talk with the AI. Not sure if things work the way I envision them though, we have no existing client using basic authentication for their AI engine. As a last resort your company could implement your own connector (using Java) for the AI engine and install it in AI Positron.
Regards,
Radu

[liyan]

Hi Radu,
Thank you again for your reply.

I was hoping your IT colleagues could give you/us a sample about how using curl one can access your AI server. In this way it would be more clear to me how the AI Positron add-on should behave.

Currently, our model uses bearer token, so the curl example is as follows:

Code: Select all

curl --location 'http://x.x.x.x:8001/v1/chat/completions' \
--header 'Content-Type: application/json' \
--header 'Authorization: Bearer xxxxxxxxxxxxxx' \
--data '{
        "model": "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B",
        "messages": [
            {
                "role": "user",
                "content": "你好，请解释一下你自己。"
            }
        ],
        "temperature": 0.1
    }'

The authentication will be switched to basic.

I would envision this to work by you no longer providing an AI Key in the preferences, then when you invoke an action you would receive an authentication dialog from Oxygen asking for username/password credentials. Then Oxygen would cache your user/password credentials and reuse them without asking you every time you talk with the AI. Not sure if things work the way I envision them though, we have no existing client using basic authentication for their AI engine.

This will be great! But as you said, this requires changes on your side. So from your perspective, it needs consideration and your side needs to weigh the benefits and costs.

As a last resort your company could implement your own connector (using Java) for the AI engine and install it in AI Positron.

I would consult my IT colleagues but I am not sure about how to install it in AI Positron. When we install Oxygen XML Editor, we have an AI positron directory, and then we can paste the jAR to that directory?

[Radu]

Hi,
Maybe for now you can test using the AI Positron Enterprise add-on with your bearer authentication (which should work out of the box with the OpenAI connector).
In the meantime we released AI Positron 5.0.0 which allows one to create an extra connector addon and connect AI Positron to their own AI engine controlling precisely how that connection works, there is a sample connector project for this here:
https://github.com/oxygenxml/oxygen-ai- ... ctor-addon

Also a what's new for the 5.0.0 Positron release:
https://www.oxygenxml.com/ai_positron/whats_new.html
Regards,
Radu

[liyan]

Thank you Radu,
Yes, I tried using the AI Positron add-on with the bearer authentication. It worked but when I tried to chat, I got the following error: Rate limit reached for RPM (rpm_rate_limit_exceeded)
I got the error after I updated the add-on. I am using oxygen xml editor 26.0. Maybe I updated the add-on, so it is not compatible?

By the way, thank you for the sample connector project!! I will study that!

[Radu]

Hi,

Can you attach a screenshot with how the error looks like in the application?
This "Rate limit reached for" error messages seems to be a message thrown by the AI engine and presented by AI Positron, so this is not something our add-on can control, you need to check your rate limitations on the server side.

Regards,
Radu

[liyan]

Hi Radu,
This is the error message. I only sent two requests and then this error appeared.

错误截图.png

[liyan]

Hi Radu,
Another question about the license. I found that our company uses a professional-floating license for Oxygen XML Editor. However, this license does not give us a free use of AI Positron. The AI Positron dialog still requires a license. Could you tell me which type of license can automatically have the AI Positron license?
Thank you very much
Regards
liyan

[Radu]

Hello Lyan,

We cannot help with the "Rate limit reached for RPM" error, please ask your system admins for help on this one, this error is reported by the AI engine and displayed by Oxygen on the client side.

About this other question:

I found that our company uses a professional-floating license for Oxygen XML Editor. However, this license does not give us a free use of AI Positron. The AI Positron dialog still requires a license. Could you tell me which type of license can automatically have the AI Positron license?

You have two options:
1) Ask our sales (sales@oxygenxml.com) how much it would cost to upgrade the "professional-floating" license to an "enterprise-floating" license which should work with the AI Positron Enterprise without any additional cost.
2) Buy a special license for the AI Positron Enterprise add-on: https://www.oxygenxml.com/ai_positron/b ... enterprise

Regards,
Radu