LISA AI Log Analyzer Failure max_output_tokens #4179

Symptom

LISA's AI log analyzer fails when attempting to use Azure OpenAI chat clients due to an invalid API parameter being passed to the Azure OpenAI service.

Root Cause

The agent-framework library incorrectly translates the max_tokens parameter to max_output_tokens when calling the Azure OpenAI Chat Completions API. The parameter max_output_tokens is not supported by Azure OpenAI's AsyncCompletions.create() method, which expects either max_tokens or max_completion_tokens.

This is a known bug in the agent-framework library:

• GitHub Issue: microsoft/agent-framework#3025
• Status: Fix in progress via PR #3037

Error Message

[INFO] ala [Magentic] Attempt 1 failed: <class 'agent_framework.azure._chat_client.AzureOpenAIChatClient'> service failed to complete the prompt: AsyncCompletions.create() got an unexpected keyword argument 'max_output_tokens'

Affected Versions

• agent-framework-core: 1.0.0b251223
• agent-framework-azure-ai: 1.0.0b251223
• openai: 2.8.1

Workaround

Instead of passing max_tokens directly to ChatAgent, the issue comment suggests using:

additional_properties={"max_completion_tokens": 8000}

to bypass the buggy translation in the framework and uses the correct Azure OpenAI API parameter.

Once agent-framework releases a fix, update to the fixed version and remove the workaround.