Skip to content

Fix: Enable Gemini support via GOOGLE_API_KEY fallback and explicit api_key injection #2805

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 4 commits into
base: main
Choose a base branch
from
+45 −9
Open

Fix: Enable Gemini support via GOOGLE_API_KEY fallback and explicit api_key injection #2805

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
21961e7
Fix: Enable Gemini support via GOOGLE_API_KEY fallback and explicit a…
MathavanSG May 10, 2025
e37dd98
Fix: finalized Gemini API key fallback and completion param injection…
MathavanSG May 10, 2025
064c550
Refactor: remove explicit api_key enforcement and redundant injection…
MathavanSG May 12, 2025
ec065a3
Merge branch 'main' into fix/gemini-langchain-support
lucasgomide Jul 2, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
54 changes: 45 additions & 9 deletions src/crewai/llm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -56,9 +56,16 @@
from crewai.utilities.exceptions.context_window_exceeding_exception import (
LLMContextLengthExceededException,
)
import logging


load_dotenv()

# Patch: Gemini key fallback
if "GEMINI_API_KEY" not in os.environ and "GOOGLE_API_KEY" in os.environ:
os.environ["GEMINI_API_KEY"] = os.environ["GOOGLE_API_KEY"]
logging.info("[CrewAI Gemini Patch] Set GEMINI_API_KEY from GOOGLE_API_KEY")


class FilteredStream(io.TextIOBase):
_lock = None
Expand Down Expand Up @@ -287,6 +294,17 @@ class AccumulatedToolArgs(BaseModel):


class LLM(BaseLLM):
"""
LLM class for handling language model interactions via LiteLLM.

Features:
- Supports multiple model providers (e.g., OpenAI, Gemini, Anthropic)
- Automatically uses GOOGLE_API_KEY if GEMINI_API_KEY is not explicitly set
- Injects the resolved API key directly into the LLM completion parameters
- Ensures compatibility with both legacy and AI Studio-style key environments
- Designed for use in CrewAI agent workflows and tool-based LLM interactions
"""

def __init__(
self,
model: str,
Expand Down Expand Up @@ -348,6 +366,15 @@ def __init__(
else:
self.stop = stop

# Fallback logic
if "GEMINI_API_KEY" in os.environ:
api_key = os.environ["GEMINI_API_KEY"]
elif "GOOGLE_API_KEY" in os.environ:
api_key = os.environ["GOOGLE_API_KEY"]
os.environ["GEMINI_API_KEY"] = api_key

self.api_key = api_key
Comment on lines +369 to +376
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You have couple of issues here

  1. If the parameter api_key was provided you have to use it even those GOOGLE/GEMINI_API_KEY have been set.
  2. move this to a dedicated method and call it on init method liek
def get_api_key(api_key: str):
   if is_gemini_model(self.model):
       return api_key or self._get_gemini_api_key_evn_vars()
   return api_key

self.api_key = def get_api_key(fallback=api_key):


self.set_callbacks(callbacks)
self.set_env_callbacks()

Expand All @@ -367,19 +394,26 @@ def _prepare_completion_params(
self,
messages: Union[str, List[Dict[str, str]]],
tools: Optional[List[dict]] = None,
**kwargs: Any,
) -> Dict[str, Any]:
"""Prepare parameters for the completion call.
"""
Prepare parameters for the LLM completion API call.

This method:
- Formats input messages for the model provider
- Accepts optional tool definitions
- Injects API key into the request (fallback to GOOGLE_API_KEY if GEMINI_API_KEY is not set)
- Merges additional keyword arguments passed to support flexibility

Args:
messages: Input messages for the LLM
tools: Optional list of tool schemas
callbacks: Optional list of callback functions
available_functions: Optional dict of available functions
messages (Union[str, List[Dict[str, str]]]): Prompt or structured messages to send to the LLM.
tools (Optional[List[dict]]): Optional tool definitions (for function calling).
**kwargs (Any): Additional optional parameters for the completion call.

Returns:
Dict[str, Any]: Parameters for the completion call
Dict[str, Any]: Final parameters dictionary to be passed to `litellm.completion(...)`.
"""
# --- 1) Format messages according to provider requirements
# Format messages
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You removed 1), but left the other section numbers like 2).
I don’t think it needs to be deleted.. rolling it back might be a better option.

if isinstance(messages, str):
messages = [{"role": "user", "content": messages}]
formatted_messages = self._format_messages_for_provider(messages)
Expand Down Expand Up @@ -411,8 +445,10 @@ def _prepare_completion_params(
**self.additional_params,
}

# Remove None values from params
return {k: v for k, v in params.items() if v is not None}
# Remove None values
params = {k: v for k, v in params.items() if v is not None}

return params

def _handle_streaming_response(
self,
Expand Down