[BUG] Model naming prefixes filtering is too strict

[Ducnguyen147]

Description

The prefix naming is too strict for our usecases. Here is the code:

crewAI/lib/crewai/src/crewai/llm.py

Line 469 in 56b6594

model_lower.startswith(prefix) for prefix in ["claude-", "anthropic."]

We have our own-deployed model so called 'anthropic--claude-...', however the CrewAI code above just filter out my models because it only supports ["claude-", "anthropic."] prefixes

Steps to Reproduce

1. Tried own Anthropic model with different naming prefixes rather than ["claude-", "anthropic."]
2. CrewAI will filter out these models which lead to incorrect behavior in the output

Expected behavior

We should think about updating this function:

crewAI/lib/crewai/src/crewai/llm.py

Line 446 in 56b6594

def _matches_provider_pattern(cls, model: str, provider: str) -> bool:

so that it can support wider range of models (e.g., own deployed model with different naming convention)

Screenshots/Code snippets

crewAI/lib/crewai/src/crewai/llm.py

Line 446 in 56b6594

def _matches_provider_pattern(cls, model: str, provider: str) -> bool:

Operating System

macOS Catalina

Python Version

3.12

crewAI Version

1.14.2

crewAI Tools Version

1.14.2

Virtual Environment

Venv

Evidence

Incorrect output behavior

Possible Solution

Update list of filtering prefixes

Additional context

None

[greysonlalonde]

Hey @Ducnguyen147 , thanks for flagging. We've had some long-running issues with the naming logic and are having internal discussions to determine a better approach. No timeline right now.

[Eashant3835]

Hey @Ducnguyen147 , thanks for flagging. We've had some long-running issues with the naming logic and are having internal discussions to determine a better approach. No timeline right now.

Hi! I have actually submitted a pull request solving this issue. Please review it and let me know what changes i can make, thank you!