Copilot Studio Agent OpenAIModelTokenLimit error

(1) Share

Report

Posted on by AD-01081101-0

Hi All,

We have created a copilot studio agent which uses "Get Emails (V3)" connector. The issue we are facing is when any email has an attachment size of more than 500 KB

than we are getting the "toomuchdatatohandle OpenAIModelTokenLimit"" error.

Even if we give the user prompt to the agent link how many emails recieved today we are getting above OpenAIModelTokenLimit error.

Any suggestions or if any one has faced the same issue and found a solution for this issue please update it here.

Regards.

Avinash.

Categories:

Bot extensibility

I have the same question (0)

All responses (3)

Answers (0)

Sort by

Suggested answer

11manish 2,286 on at

Like (0)

Report
Copy link

Link copied!
Do NOT allow Microsoft Copilot Studio to process raw Outlook email payloads directly.

The correct architecture is:

Use Power Automate as a preprocessing layer

Strip attachments/body HTML

Return lightweight structured JSON only

Process attachments separately if needed

This is the standard and scalable solution for avoiding OpenAIModelTokenLimit issues.

Was this reply helpful? Yes No
Nivedipa-MSFT Microsoft Employee on at

Like (0)

Report
Copy link

Link copied!
Hello AD-01081101-0 ,
Root cause: The Get Emails (V3) action returns the full email body along with base64 attachments, which exceeds the model's context window. This large output remains in context, so even simple queries like "how many emails today" can hit the limit.

Recommended steps:

In the Get Emails action, set Include Attachments to No. This is the most effective change.

Set Top to a smaller value, such as 10–25.

Use $select to retrieve only essential fields: subject, from, receivedDateTime, and bodyPreview (not the full body).

Apply $filter server-side to limit by date range.

Instead of passing the entire connector output to the LLM, extract only the required fields into a variable and use that.

In Settings → Generative AI, reduce the number of conversation history turns (e.g., to 3).

If possible, switch to GPT-4o / GPT-4 Turbo (128K).

Setting Include Attachments to No and using $select for minimal fields will resolve this issue in about 95% of cases.

Was this reply helpful? Yes No
Suggested answer

Haque 2,952 on at

Like (1)

Report
Copy link

Link copied!

Hi @AD-01081101-0,

Exclude attachments from the prompt or set the low limit: Do not include attachment content or large base64 data in the prompt sent to OpenAI if not needed or at least custom your prompt to deal with less than 20/30/50 MB.

I am sure some clues I tried to give. If these clues help to resolve the issue brought you by here, please don't forget to check the box Does this answer your question? At the same time, I am pretty sure you have liked the response!

Was this reply helpful? Yes No