Jens Olaf Koch 

Jens Olaf Koch said:

Is every API request to the OpenAI models independent from all prior requests and all surrounding segments within the current source file? I guess so because the API endpoints are completion-endpoints?

Yes.

Jens Olaf Koch said:

If my guess is correct, there would be no context-awareness in OpenAI translator yet?

Correct

Jens Olaf Koch said:

So maybe it could be possible to add a feature which would allow us to transfer a complete source text of a file as general context and then (as long as we are in an editor session) transmit the individual requests as part of an ongoing "chat" (still via API calls)? That way we could (theoretically) have the best of both worlds: context awareness and the segment-wise functionality.

Certainly this is possible and we have discussed this.  It would come at a cost financially and in terms of performance but it could be done.  We thought about whole document, and also about setting parameters before and after the active segment but for now opted not to do this since the performance would not be good and there are complexities around having to keep the context live for the duration of the project, when you close and restart again etc.

Nothing that probably couldn't be resolved, but all at a price and for this phase we took the view we should concentrate on other features.

 Jens Olaf Koch 

Jens Olaf Koch said:

Is every API request to the OpenAI models independent from all prior requests and all surrounding segments within the current source file? I guess so because the API endpoints are completion-endpoints?

Yes.

Jens Olaf Koch said:

If my guess is correct, there would be no context-awareness in OpenAI translator yet?

Correct

Jens Olaf Koch said:

So maybe it could be possible to add a feature which would allow us to transfer a complete source text of a file as general context and then (as long as we are in an editor session) transmit the individual requests as part of an ongoing "chat" (still via API calls)? That way we could (theoretically) have the best of both worlds: context awareness and the segment-wise functionality.

Certainly this is possible and we have discussed this.  It would come at a cost financially and in terms of performance but it could be done.  We thought about whole document, and also about setting parameters before and after the active segment but for now opted not to do this since the performance would not be good and there are complexities around having to keep the context live for the duration of the project, when you close and restart again etc.

Nothing that probably couldn't be resolved, but all at a price and for this phase we took the view we should concentrate on other features.

 Paul  Thank you for the detailed explanation!