Hey again team! one question... if i want to give ...

# koog-agentic-framework

[Eduardo Ruesta]

Hey again team! one question... if i want to give the agent .pdf document to learn and answer question about the information inside it the .pdf. The best way to do it is in the prompt using attachments?

[Vadim Briliantov]

Hi! It depends on the model you are willing to use. Not all models support documents, and some that do might be more expensive. So it’s basically a trade-off between using some external tools for pdf->text extraction and then providing the text to a cheaper model VS using a smaeter model without any extra tools/libraries.

For the start I would suggest doing what you described — just add pdf as attachment

[Eduardo Ruesta]

okay! and for example if i want to use OpenAIModels.Chat.GPT4o... can i read documents? and this is the way to do it:

attachments {
    attachment(
        Attachment.File(
            content = AttachmentContent.Binary.Bytes(readPdfAsBytes("/Users/eduardoruesta/Workplace/SafePal/uso_pantalla.pdf")),
            format = "pdf",
            mimeType = "application/pdf",
        )
    )
    attachment(
        Attachment.File(
            content = AttachmentContent.Binary.Bytes(readPdfAsBytes("/Users/eduardoruesta/Workplace/SafePal/bebes_ninos_pantalla.pdf")),
            format = "pdf", 
            mimeType = "application/pdf",
        )
    )
}

[Eduardo Ruesta]

i have an error: Model gpt-4o does not support files

[Vadim Briliantov]

What if you try something like

val MyGPT4o = with(OpenAIModels.Chat.GPT4o) {
   copy(capabilities = capabilities + LLMCapability.Document)
}

as was described above in the thread?

[Eduardo Ruesta]

yea!! worked. now facing another issue haha

[Eduardo Ruesta]

Error: Error from OpenAl API: -1 Unknown
Status Code:
"error": م
"message": "Missing required parameter:
messages[1].content[1].file.fle_id".",
type": "invalid_request_error"
param": "messages[1].content[1].file.file_id",
"code": "missing_required_parameter
)

[Eduardo Ruesta]

I looked up the problem, and it says I need to upload the files to OpenAI first. Is that correct? Or is there a way to get the model to read my PDFs?

[Pavel Gorgulov]

it says I need to upload the files to OpenAI first

It’s expected to work like that, but OpenAI allowed sending files directly, and file_id is optional field

I’ll look into what the issue is

[Pavel Gorgulov]

Unfortunately, OpenAI now really requires using the file api, need to upload file and then specify file id
Even though the documentation says that file_id is optional

there’s no workaround for this using attachments. The only option is to read the text and include it as part of the prompt

[Eduardo Ruesta]

Yes, i saw the same. Thats why i moved to Gemini and worked! Thanks 🙏🏽

[Pavel Gorgulov]

Possibly, this is related to issues on openai side and the current release of gpt-5. A month ago, pdf files were working with openai.

Nevertheless, I’ve created an issue to use the File API - KG-215

[Eduardo Ruesta]

niceee!! thanks for that kodee happy