Skip to content

Automate browser based workflows with AI

refactor gemini reasoning effor (#3292)

LawyZhenge0043d002ce3
diff --git a/skyvern/config.py b/skyvern/config.py
index 7176b67c8..d826bda59 100644
--- a/skyvern/config.py
+++ b/skyvern/config.py
@@ -245,7 +245,8 @@ class Settings(BaseSettings):
# GEMINI
GEMINI_API_KEY: str | None = None
- GEMINI_REASONING_EFFORT: str | None = None
+ GEMINI_INCLUDE_THOUGHT: bool = False
+ GEMINI_THINKING_BUDGET: int | None = None
# VERTEX_AI
VERTEX_CREDENTIALS: str | None = None
diff --git a/skyvern/forge/sdk/api/llm/config_registry.py b/skyvern/forge/sdk/api/llm/config_registry.py
index 8b3ce2f3e..44d155f69 100644
--- a/skyvern/forge/sdk/api/llm/config_registry.py
+++ b/skyvern/forge/sdk/api/llm/config_registry.py
@@ -787,7 +787,12 @@ if settings.ENABLE_GEMINI:
supports_vision=True,
add_assistant_prefix=False,
max_completion_tokens=65536,
- reasoning_effort=settings.GEMINI_REASONING_EFFORT,
+ litellm_params=LiteLLMParams(
+ thinking={
+ "budget_tokens": settings.GEMINI_THINKING_BUDGET,
+ "type": "enabled" if settings.GEMINI_INCLUDE_THOUGHT else None,
+ },
+ ),
),
)
LLMConfigRegistry.register_config(
@@ -798,7 +803,12 @@ if settings.ENABLE_GEMINI:
supports_vision=True,
add_assistant_prefix=False,
max_completion_tokens=65536,
- reasoning_effort=settings.GEMINI_REASONING_EFFORT,
+ litellm_params=LiteLLMParams(
+ thinking={
+ "budget_tokens": settings.GEMINI_THINKING_BUDGET,
+ "type": "enabled" if settings.GEMINI_INCLUDE_THOUGHT else None,
+ },
+ ),
),
)
LLMConfigRegistry.register_config(
@@ -809,7 +819,12 @@ if settings.ENABLE_GEMINI:
supports_vision=True,
add_assistant_prefix=False,
max_completion_tokens=65536,
- reasoning_effort=settings.GEMINI_REASONING_EFFORT,
+ litellm_params=LiteLLMParams(
+ thinking={
+ "budget_tokens": settings.GEMINI_THINKING_BUDGET,
+ "type": "enabled" if settings.GEMINI_INCLUDE_THOUGHT else None,
+ },
+ ),
),
)
LLMConfigRegistry.register_config(
@@ -820,7 +835,12 @@ if settings.ENABLE_GEMINI:
supports_vision=True,
add_assistant_prefix=False,
max_completion_tokens=65536,
- reasoning_effort=settings.GEMINI_REASONING_EFFORT,
+ litellm_params=LiteLLMParams(
+ thinking={
+ "budget_tokens": settings.GEMINI_THINKING_BUDGET,
+ "type": "enabled" if settings.GEMINI_INCLUDE_THOUGHT else None,
+ },
+ ),
),
)
LLMConfigRegistry.register_config(
@@ -831,7 +851,12 @@ if settings.ENABLE_GEMINI:
supports_vision=True,
add_assistant_prefix=False,
max_completion_tokens=65536,
- reasoning_effort=settings.GEMINI_REASONING_EFFORT,
+ litellm_params=LiteLLMParams(
+ thinking={
+ "budget_tokens": settings.GEMINI_THINKING_BUDGET,
+ "type": "enabled" if settings.GEMINI_INCLUDE_THOUGHT else None,
+ },
+ ),
),
)
@@ -1027,8 +1052,11 @@ if settings.ENABLE_VERTEX_AI and settings.VERTEX_CREDENTIALS:
vertex_credentials=settings.VERTEX_CREDENTIALS,
api_base=f"{api_base}/gemini-2.5-pro" if api_base else None,
vertex_location=settings.VERTEX_LOCATION,
+ thinking={
+ "budget_tokens": settings.GEMINI_THINKING_BUDGET,
+ "type": "enabled" if settings.GEMINI_INCLUDE_THOUGHT else None,
+ },
),
- reasoning_effort=settings.GEMINI_REASONING_EFFORT,
),
)
LLMConfigRegistry.register_config(
@@ -1043,8 +1071,11 @@ if settings.ENABLE_VERTEX_AI and settings.VERTEX_CREDENTIALS:
vertex_credentials=settings.VERTEX_CREDENTIALS,
api_base=f"{api_base}/gemini-2.5-pro-preview-05-06" if api_base else None,
vertex_location=settings.VERTEX_LOCATION,
+ thinking={
+ "budget_tokens": settings.GEMINI_THINKING_BUDGET,
+ "type": "enabled" if settings.GEMINI_INCLUDE_THOUGHT else None,
+ },
),
- reasoning_effort=settings.GEMINI_REASONING_EFFORT,
),
)
LLMConfigRegistry.register_config(
@@ -1059,8 +1090,11 @@ if settings.ENABLE_VERTEX_AI and settings.VERTEX_CREDENTIALS:
vertex_credentials=settings.VERTEX_CREDENTIALS,
api_base=f"{api_base}/gemini-2.5-flash" if api_base else None,
vertex_location=settings.VERTEX_LOCATION,
+ thinking={
+ "budget_tokens": settings.GEMINI_THINKING_BUDGET,
+ "type": "enabled" if settings.GEMINI_INCLUDE_THOUGHT else None,
+ },
),
- reasoning_effort=settings.GEMINI_REASONING_EFFORT,
),
)
LLMConfigRegistry.register_config(
@@ -1075,8 +1109,11 @@ if settings.ENABLE_VERTEX_AI and settings.VERTEX_CREDENTIALS:
vertex_credentials=settings.VERTEX_CREDENTIALS,
api_base=f"{api_base}/gemini-2.5-flash-preview-05-20" if api_base else None,
vertex_location=settings.VERTEX_LOCATION,
+ thinking={
+ "budget_tokens": settings.GEMINI_THINKING_BUDGET,
+ "type": "enabled" if settings.GEMINI_INCLUDE_THOUGHT else None,
+ },
),
- reasoning_effort=settings.GEMINI_REASONING_EFFORT,
),
)
LLMConfigRegistry.register_config(
@@ -1091,8 +1128,11 @@ if settings.ENABLE_VERTEX_AI and settings.VERTEX_CREDENTIALS:
vertex_credentials=settings.VERTEX_CREDENTIALS,
api_base=f"{api_base}/gemini-2.5-flash-preview-04-17" if api_base else None,
vertex_location=settings.VERTEX_LOCATION,
+ thinking={
+ "budget_tokens": settings.GEMINI_THINKING_BUDGET,
+ "type": "enabled" if settings.GEMINI_INCLUDE_THOUGHT else None,
+ },
),
- reasoning_effort=settings.GEMINI_REASONING_EFFORT,
),
)
LLMConfigRegistry.register_config(
@@ -1107,8 +1147,11 @@ if settings.ENABLE_VERTEX_AI and settings.VERTEX_CREDENTIALS:
vertex_credentials=settings.VERTEX_CREDENTIALS,
api_base=f"{api_base}/gemini-2.5-flash-preview-05-20" if api_base else None,
vertex_location=settings.VERTEX_LOCATION,
+ thinking={
+ "budget_tokens": settings.GEMINI_THINKING_BUDGET,
+ "type": "enabled" if settings.GEMINI_INCLUDE_THOUGHT else None,
+ },
),
- reasoning_effort=settings.GEMINI_REASONING_EFFORT,
),
)
LLMConfigRegistry.register_config(
diff --git a/skyvern/forge/sdk/api/llm/models.py b/skyvern/forge/sdk/api/llm/models.py
index 9d231c2a5..1e4ff6ad3 100644
--- a/skyvern/forge/sdk/api/llm/models.py
+++ b/skyvern/forge/sdk/api/llm/models.py
@@ -16,6 +16,7 @@ class LiteLLMParams(TypedDict, total=False):
model_info: dict[str, Any] | None
vertex_credentials: str | None
vertex_location: str | None
+ thinking: dict[str, Any] | None
@dataclass(frozen=True)