Skip to content

Automate browser based workflows with AI

Fix openai flex usage (#4141)

Suchintand6aed0d0be7e
diff --git a/skyvern/forge/sdk/api/llm/api_handler_factory.py b/skyvern/forge/sdk/api/llm/api_handler_factory.py
index 46655c2dd..042e6e534 100644
--- a/skyvern/forge/sdk/api/llm/api_handler_factory.py
+++ b/skyvern/forge/sdk/api/llm/api_handler_factory.py
@@ -653,6 +653,9 @@ class LLMAPIHandlerFactory:
if llm_config.litellm_params: # type: ignore
active_parameters.update(llm_config.litellm_params) # type: ignore
+ if "timeout" not in active_parameters:
+ active_parameters["timeout"] = settings.LLM_CONFIG_TIMEOUT
+
# Apply thinking budget optimization if settings are available
if (
LLMAPIHandlerFactory._thinking_budget_settings
@@ -773,13 +776,11 @@ class LLMAPIHandlerFactory:
t_llm_request = time.perf_counter()
try:
- # TODO (kerem): add a timeout to this call
# TODO (kerem): add a retry mechanism to this call (acompletion_with_retries)
# TODO (kerem): use litellm fallbacks? https://litellm.vercel.app/docs/tutorials/fallbacks#how-does-completion_with_fallbacks-work
response = await litellm.acompletion(
model=model_name,
messages=messages,
- timeout=settings.LLM_CONFIG_TIMEOUT,
drop_params=True, # Drop unsupported parameters gracefully
**active_parameters,
)
diff --git a/skyvern/forge/sdk/api/llm/models.py b/skyvern/forge/sdk/api/llm/models.py
index aff891ce8..45274c57d 100644
--- a/skyvern/forge/sdk/api/llm/models.py
+++ b/skyvern/forge/sdk/api/llm/models.py
@@ -19,6 +19,7 @@ class LiteLLMParams(TypedDict, total=False):
vertex_location: str | None
thinking: dict[str, Any] | None
service_tier: str | None
+ timeout: float | None
@dataclass(frozen=True)