[BUG] Run azure_openai_neo4j sample code bug failed.

[nadazero]

Bug Description

Run sample code azure_openai_neo4j.py in project but failed.

Steps to Reproduce

Run sample code azure_openai_neo4j.py in project but failed.

Expected Behavior

Run sample code azure_openai_neo4j.py normally

Actual Behavior

Run sample code azure_openai_neo4j.py in project but failed.

Environment

• Graphiti Version: [0.28.1]
• Python Version: [3.11.7]
• Operating System: [Windows10]
• Database Backend: [Neo4j 5.26.2]
• LLM Provider & Model: [OpenAI gpt-5]

Installation Method

Error Messages/Traceback

Request: POST https://xxx.openai.azure.com/openai/v1/responses "HTTP/1.1 200 OK"
2026-03-04 16:03:39 - graphiti_core.llm_client.openai_base_client - WARNING - Retrying after application error (attempt 1/2): not enough values to unpack (expected 3, got 1)
2026-03-04 16:03:49 - httpx - INFO - HTTP Request: POST https://xxx.openai.azure.com/openai/v1/responses "HTTP/1.1 200 OK"
2026-03-04 16:03:49 - graphiti_core.llm_client.openai_base_client - WARNING - Retrying after application error (attempt 2/2): not enough values to unpack (expected 3, got 1)
2026-03-04 16:03:55 - httpx - INFO - HTTP Request: POST https://xxx.openai.azure.com/openai/v1/responses "HTTP/1.1 200 OK"
2026-03-04 16:03:55 - graphiti_core.llm_client.openai_base_client - ERROR - Max retries (2) exceeded. Last error: not enough values to unpack (expected 3, got 1)

Configuration

Additional Context

Possible Solution

Here is AI's analysis:
Cause of error:
Base class method signature mismatch: BaseOpenAIClient._handle_structured_response() returns tuple[dict, int, int] (3 values).
Azure subclass overridden method: AzureOpenAILLMClient._handle_structured_response() only returns dict (1 value).
Caller expects 3 values: The _generate_response() method calls _handle_structured_response() on line 181 and expects a 3-value tuple to be returned.
Unpacking failed on line 245: response, input_tokens, output_tokens = await self._generate_response(...)
Root cause: The Azure OpenAI client overrode the _handle_structured_response method but failed to correctly extract and return the token count information.