[BUG]

[rschwegler]

Description

Running crewai on docker with Ollama on host configuration generation is wrong.

Steps to Reproduce

Setup of crewai in a docker container running ubuntu. Ollama installed on the host (MacOS).
crewai create crew demo

asks for llm where I selected 5. ollama and 1. ollama/llama3.1 as model. This step produced the .env file with the content:

MODEL=ollama/llama3.1 API_BASE=http://localhost:11434

The correct url is host.docker.internal which I changed and also adopted my model.

Following error when starting crewai run:

ERROR:root:LiteLLM call failed: litellm.APIConnectionError: OllamaException - [Errno 99] Cannot assign requested address

Expected behavior

Run the example printing output of LLM.

Screenshots/Code snippets

Operating System

Ubuntu 24.04

Python Version

3.12

crewAI Version

crewai version: 0.86.0

crewAI Tools Version

n.A.

Virtual Environment

Venv

Evidence

see screenshot above

Possible Solution

Instead of setting API_BASE set OPENAI_API_BASE=http://host.docker.internal:11434 in .env file.
Alternatively, add in crewai/agent.py: 167 the check for API_BASE:

167             api_base = (
168                 os.environ.get("OPENAI_API_BASE")
169                 or os.environ.get( "OPENAI_BASE_URL")
170                 or os.environ.get( "API_BASE")
171             )

Additional context

I would prefer the code addition from the possible solution, because semantically it is not the OPENAI_BASE_URL ;-)

[imsharukh1994]

Solution 1: Correct the .env File

Update the .env file as follows:

MODEL=ollama/llama3.1
OPENAI_API_BASE=http://host.docker.internal:11434

This ensures compatibility since crewai is expecting the OPENAI_API_BASE variable.

Solution 2: Code Update in crewai/agent.py

You can modify the agent.py file to include a fallback for API_BASE. Update the code around line 167 to:

api_base = (
    os.environ.get("OPENAI_API_BASE")
    or os.environ.get("OPENAI_BASE_URL")
    or os.environ.get("API_BASE")  # Added support for API_BASE
)

This change ensures that API_BASE is recognized as a valid variable without needing to rename it in the .env file.