Copilot Studio Agent in Teams stuck after "Request body too large" error

The important distinction is between the size of the uploaded PDF and the size of the request ultimately sent to the model. A 7.7 MB PDF can potentially generate a much larger amount of contextual data after processing, particularly when the document contains many images and requires OCR/content extraction. The resulting request may include extracted text, image-related content, attachment metadata, conversation history, and other agent context. Therefore, the 30,000,000 bytes limit in the error should not be interpreted as a 30 MB limit on the original PDF itself. It is a limit on the request body being sent to the OpenAI/model endpoint.

The more interesting part of your case is that even hello fails after the PDF upload. If the problem were simply that the PDF itself exceeded a file-upload limit, you would normally expect the subsequent hello message to work. The fact that unrelated messages continue to produce:

Request body too large ContextTokenLimitExceeded

suggests that some of the conversation/session context created by the attachment may still be included when processing subsequent messages.

This is particularly relevant to Teams because Teams supports persistent conversations. In Copilot Studio, resetting the conversation and clearing the conversation history are not necessarily the same thing. The normal reset behavior can reset variables/state without necessarily removing all conversation history that is being supplied to the planner/model.

The fact that the same agent works correctly in the Microsoft Copilot app but fails in Teams is another strong indication that the agent itself is probably not fundamentally broken. It points more toward a difference in how the Teams channel is maintaining or reconstructing the conversation/session context.

Test this in the following order:

1. In the affected Teams conversation, send start over and then send hello.
2. If it still fails, create a completely new Teams conversation/chat with the same agent.
3. Send hello in the new conversation before uploading any file.
4. If that works, try a small PDF first and then the original PDF.
5. If even a brand-new Teams conversation fails with hello, while the same agent continues to work in Microsoft Copilot, this will be considered as a strong evidence of a Teams/Copilot Studio service-side or session-state issue.

Thank you for the detailed explanation.

However, the documented recovery actions do not resolve the issue in my case.

Additional findings:

• The agent continues to work correctly in the Microsoft Copilot app.
• Other users can use the Teams agent without any problem.
• A colleague reproduced the issue after uploading the same large PDF containing many images. His Teams conversation is now also permanently broken.
• Users who did not upload the PDF are not affected.

I have already tested:

• Deleting conversation history
• Removing and reinstalling the agent in Teams
• Republishing the agent
• Using /foundry_new_preview

None of these actions changed the behavior.

What is particularly surprising is that /foundry_new_preview itself returns the same Request body too large error, which suggests the reset command may never actually be executed.

It appears that once the large PDF is uploaded, the Teams conversation enters a permanently broken state where every subsequent message (even a simple "hello") immediately returns the same error.

Based on the behavior observed, this does not seem to be an agent configuration issue because:

• The same agent works correctly in Copilot.
• Other users can still use the agent.
• Only conversations that previously received the large PDF become affected.

Is there any documented way to completely reset or purge the server-side Teams conversation/session associated with a Copilot Studio agent?

Also, should the conversation ID change after using /foundry_new_preview or after deleting the conversation history? In my case, the conversation appears to remain stuck in the same state.

This currently looks like a Teams conversation/session corruption issue rather than a normal context limit error.Thank you for the detailed explanation.

However, the documented recovery actions do not resolve the issue in my case.

Additional findings:

• The agent continues to work correctly in the Microsoft Copilot app.
• Other users can use the Teams agent without any problem.
• A colleague reproduced the issue after uploading the same large PDF containing many images. His Teams conversation is now also permanently broken.
• Users who did not upload the PDF are not affected.

I have already tested:

• Deleting conversation history
• Removing and reinstalling the agent in Teams
• Republishing the agent
• Using /foundry_new_preview

None of these actions changed the behavior.

What is particularly surprising is that /foundry_new_preview itself returns the same Request body too large error, which suggests the reset command may never actually be executed.

It appears that once the large PDF is uploaded, the Teams conversation enters a permanently broken state where every subsequent message (even a simple "hello") immediately returns the same error.

Based on the behavior observed, this does not seem to be an agent configuration issue because:

• The same agent works correctly in Copilot.
• Other users can still use the agent.
• Only conversations that previously received the large PDF become affected.

Is there any documented way to completely reset or purge the server-side Teams conversation/session associated with a Copilot Studio agent?

Also, should the conversation ID change after using /foundry_new_preview or after deleting the conversation history? In my case, the conversation appears to remain stuck in the same state.

This currently looks like a Teams conversation/session corruption issue rather than a normal context limit error.