For example certain private data from a pdf file could be filtered and not be sent to Open AI or another AI for the analysis of the content which is allowed to be used.
Redacted or completely avoided would solve my privacy requirements. Right now I run a python script to do that for me. Either by text input or by selecting or drawing blocks around parts of a pdf with the mouse which I want to filter out before the data gets sent to Open AI.
Also a positive selection could do a similar job. That would even reduce the amount of tokens which will be used. Many times the part of interest within a pdf is only a smaller part of the pdf content. In the end I would not send the full original pdf content for analysis to a cloud based AI.
Local AI might do the job some day. My experiences so far (Mistral) were not sufficiently good.
Saggut
Team•
Jul 28, 2025
Good one! Been thinking about this for a while now as well. Have been considering a redaction feature - is that something would be even better? Or does not sending at all align with you needs better?
Log in to comment and vote
Comments2
Martin64
Jul 28, 2025
Redacted or completely avoided would solve my privacy requirements. Right now I run a python script to do that for me. Either by text input or by selecting or drawing blocks around parts of a pdf with the mouse which I want to filter out before the data gets sent to Open AI.
Also a positive selection could do a similar job. That would even reduce the amount of tokens which will be used. Many times the part of interest within a pdf is only a smaller part of the pdf content. In the end I would not send the full original pdf content for analysis to a cloud based AI.
Local AI might do the job some day. My experiences so far (Mistral) were not sufficiently good.
Saggut
Jul 28, 2025
Good one! Been thinking about this for a while now as well. Have been considering a redaction feature - is that something would be even better? Or does not sending at all align with you needs better?