Power Platform Community Forum Thread Details

Olá, pessoal!
Estou desenvolvendo um agente no Copilot Studio e estou enfrentando dificuldades relacionadas à performance e processamento durante a leitura de arquivos, ainda na fase de testes.
Especificamente:

O agente não está conseguindo interpretar corretamente o conteúdo dos arquivos (ex: PDFs/Excel/Documentos).

Em alguns casos, o processamento parece lento ou incompleto.
Não está claro se a limitação é de configuração, tamanho do arquivo, ou do próprio modelo.

Alguém já passou por esse tipo de problema?

Existe alguma boa prática para melhorar a leitura e o processamento de arquivos?

Há limites recomendados de tamanho ou formato?

É melhor utilizar algum conector ou abordagem específica (ex: armazenamento intermediário, chunking dos dados, etc.)?

Agradeço qualquer dica ou direcionamento!

Categories:

Copilot Studio pre-built agents/templates

hello,

as you have probably notice, most question are in english :) you'r company copilot could help you to translate before sending. It help a lot to find people who can help you with you'r question to use the common language of this place :)

i translated you'r question for next people

"""""

Hello everyone!

I’m currently developing an agent in Copilot Studio and I’m facing some difficulties with performance and processing when reading files, still during the testing phase.

More specifically:

The agent is not correctly interpreting the content of files, such as PDFs, Excel files, or documents.

In some cases, the processing seems slow or incomplete.

It’s not clear whether the limitation comes from the configuration, the file size, or the model itself.

Has anyone already faced this kind of issue?

Are there any best practices to improve file reading and processing?

Are there any recommended limits in terms of file size or format?

Is it better to use a specific connector or approach, such as intermediate storage, data chunking, etc.?

Thanks in advance for any advice or guidance!"

""""

Without knowing what you did i could only give basic recommandation :

First of all the model is important : recommand GPT 5 chat, wich know how to use pdf, etc

then in settings keep this one , in very old tutorial it was said to be "no" but it's very old way. it help a lot the agent to choose the good knowledge base

at the bottom of this page you have this one : keep yes, in old tutorial it said "no" but... again old tutorial : it change a lot in quality

Then for PDF : many company "pdf" are not real pdf but just "image scan" in pdf : it depend of the quality of the pdf. Real PDF have absolutly no problem for content and quality for nearly a year.

There are some limitation about size, and number of page (above 50 pages in a document it start to drop in quality, like in most public AI)

I deeply recommand you to folow this tutorial before trying to make "pro agent" : it will give you a lot of best practise and advice on how to :) https://microsoft.github.io/agent-academy/recruit/

Last "intermediate storage, data chunking" could be done for very complexe document, but it's when we speak about big industrial document hunderd of pages, with schematics etc. In rare case for very complexe business document in industrial context, copilot studo was not enough but very rare case.

I hope, thoose informations could help you :) if yes please mark the answer as verified in green, it's important for the search engine and the community.