Agent from Azure AI Foundry Works in Playground but Fails in Microsoft Teams via Bot Framework – Context Loss, Formatting, and Vector DB Retrieval Issues

We’ve built an AI Agent using Azure AI Foundry, and it works perfectly within the Foundry Playground. However, when deployed to Microsoft Teams via the Bot Framework, we are encountering several major issues:

❗ Issues in Microsoft Teams (via Bot Framework):

Context Loss Across Turns

• The agent forgets earlier user inputs and keeps asking the same question repeatedly, even in a single-threaded conversation.

Occasional Wrong or Hallucinated Answers

• Answers are sometimes inaccurate, incomplete, or irrelevant—despite the same input returning correct results in Foundry Playground.
• Unprofessional and Inconsistent Formatting

Teams often displays:

Large bold text blocks

Missing or broken key → value formatting

No markdown rendering or unexpected layout issues

In Playground:

The agent maintains proper context

Vector DB search works seamlessly

Formatting follows markdown with clean key-value presentation

Looking for Guidance On:

Are there known compatibility or memory issues when using Azure AI Foundry with Bot Framework in Teams?

Any recommendations to persist context/session state across messages in Teams chats?

What are the best practices for formatting responses in Teams (e.g., markdown support or adaptive cards)?

Is there a fix or workaround to restore vector-based retrieval in Foundry post update?

Appreciate any help, documentation pointers, or workaround suggestions.

Thanks,