Agent() Parameter Name Inconsistency (llm vs LLM)

[srishrachamalla7]

Description

There is a bug in the Agent() class where the parameter name for specifying the language model (llm) is case-sensitive, leading to unexpected behavior:

1. Using llm=llm works correctly with a custom LLM (e.g., Ollama).
2. Using LLM=llm causes the system to fallback to OpenAI's LLM, prompting for an API key (liteLLM error).

This inconsistency causes confusion and may lead to unintended fallback behavior.

Steps to Reproduce

1. Define a custom LLM instance:

llm = LLM(
model="ollama/phi3:latest",
base_url="http://localhost:11434"
)

2. Create an Agent() using:

•   Correct Usage: llm=llm (works as expected).
  

•   Incorrect Usage: LLM=llm (causes fallback to OpenAI).
  

3. Observe the behavior.

Expected behavior

The Agent() class should:

1. Enforce a consistent parameter naming convention.
2. Throw an error or warning if an unsupported parameter (LLM) is passed.

Screenshots/Code snippets

• When LLM=llm is used, the system ignores the custom LLM and defaults to OpenAI's LLM, prompting for an API key.

Operating System

Windows 11

Python Version

3.12

crewAI Version

0.86.0

crewAI Tools Version

0.17.0

Virtual Environment

Venv

Evidence

With LLM used

the code:

terminal output:

With llm used

The code:

terminal output:
** The code is working

Possible Solution

The issue can be resolved by:

1. Standardizing the parameter name in the Agent() class to llm across all usage scenarios.
2. Adding a validation step in the Agent() initialization to raise an error or warning if an unsupported parameter like LLM is used.
3. Updating the documentation to clearly state that llm is the correct parameter name.

Additional context

This issue is critical for users integrating custom LLMs like Ollama and can lead to unnecessary fallback to OpenAI if not addressed.