CrewAI with a custom OpenAI-compatible backend

1. 01

   Create the LLM

   CrewAI's LLM wrapper accepts base_url and model, and speaks the chat-completions protocol.

   python

   from crewai import Agent, Crew, Task, LLM
   
   llm = LLM(
       model="openai/gpt-6-sol",
       base_url="https://uncensoredai.dev/v1",
       api_key="sk-ua-XXXX-XXXX-XXXX-XXXX",
   )
   
   researcher = Agent(
       role="Researcher",
       goal="Find primary sources",
       backstory="Meticulous and fast.",
       llm=llm,
   )

2. 02

   Assign tasks

   Tasks inherit the agent's model. Mix tiers by giving a second agent a cheaper LLM.

   crew

   task = Task(
       description="Summarise the findings in five bullets.",
       expected_output="Five bullet points.",
       agent=researcher,
   )
   Crew(agents=[researcher], tasks=[task]).kickoff()

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