AutoGen with a custom OpenAI-compatible model

1. 01

   Create the client

   AutoGen v0.4 takes a model client. Point it at the gateway and declare the model info the agent needs.

   python

   from autogen_ext.models.openai import OpenAIChatCompletionClient
   
   client = OpenAIChatCompletionClient(
       model="deepseek-v4.1-flash",
       base_url="https://uncensoredai.dev/v1",
       api_key="sk-ua-XXXX-XXXX-XXXX-XXXX",
       model_info={
           "vision": False,
           "function_calling": True,
           "json_output": True,
           "family": "unknown",
       },
   )

2. 02

   Run an assistant

   The assistant agent is model-agnostic once the client is configured.

   agent

   from autogen_agentchat.agents import AssistantAgent
   
   assistant = AssistantAgent(name="assistant", model_client=client)

Commands and configuration below are shown exactly as the tool expects them, so they stay in English.