Skip to content

Automate browser based workflows with AI

improve chain click logic (#4166)

LawyZhenga02ccee7699a
diff --git a/skyvern/library/embedded_server_factory.py b/skyvern/library/embedded_server_factory.py
index 0e9f73cd8..cc4b38cab 100644
--- a/skyvern/library/embedded_server_factory.py
+++ b/skyvern/library/embedded_server_factory.py
@@ -1,15 +1,9 @@
-from typing import Any
-
import httpx
from httpx import ASGITransport
-from skyvern.forge.sdk.api.llm.config_registry import LLMConfigRegistry
-from skyvern.forge.sdk.api.llm.models import LLMConfig, LLMRouterConfig
-
def create_embedded_server(
- llm_config: LLMRouterConfig | LLMConfig | None = None,
- settings_overrides: dict[str, Any] | None = None,
+ openai_api_key: str | None,
) -> httpx.AsyncClient:
class EmbeddedServerTransport(httpx.AsyncBaseTransport):
def __init__(self) -> None:
@@ -21,20 +15,8 @@ def create_embedded_server(
settings.BROWSER_LOGS_ENABLED = False
- if llm_config:
- LLMConfigRegistry.register_config(
- "CUSTOM_LLM",
- llm_config,
- )
- settings.LLM_KEY = "CUSTOM_LLM"
-
- # Apply custom settings overrides
- if settings_overrides:
- for key, value in settings_overrides.items():
- if hasattr(settings, key):
- setattr(settings, key, value)
- else:
- raise ValueError(f"Invalid setting: {key}")
+ if openai_api_key:
+ settings.OPENAI_API_KEY = openai_api_key
from skyvern.forge.api_app import create_api_app # noqa: PLC0415
diff --git a/skyvern/library/skyvern.py b/skyvern/library/skyvern.py
index 670fd1c8f..f71108869 100644
--- a/skyvern/library/skyvern.py
+++ b/skyvern/library/skyvern.py
@@ -9,7 +9,6 @@ from playwright.async_api import Playwright, async_playwright
from skyvern.client import AsyncSkyvern, BrowserSessionResponse, SkyvernEnvironment
from skyvern.client.types.task_run_response import TaskRunResponse
from skyvern.client.types.workflow_run_response import WorkflowRunResponse
-from skyvern.forge.sdk.api.llm.models import LLMConfig, LLMRouterConfig
from skyvern.library.constants import DEFAULT_AGENT_HEARTBEAT_INTERVAL, DEFAULT_AGENT_TIMEOUT, DEFAULT_CDP_PORT
from skyvern.library.embedded_server_factory import create_embedded_server
from skyvern.library.skyvern_browser import SkyvernBrowser
@@ -107,55 +106,15 @@ class Skyvern(AsyncSkyvern):
def __init__(
self,
*,
- llm_config: LLMRouterConfig | LLMConfig | None = None,
- settings: dict[str, Any] | None = None,
+ openai_api_key: str | None = None,
) -> None:
"""Embedded mode: Run Skyvern locally in-process.
- Prerequisites:
- Run `skyvern quickstart` first to set up your local environment and create a .env file.
+ To use this mode, run `skyvern quickstart` first.
Args:
- llm_config: Optional custom LLM configuration (LLMConfig or LLMRouterConfig).
- If provided, this will be registered as "CUSTOM_LLM" and used as the primary LLM,
- overriding the LLM_KEY setting from your .env file.
- If not provided, uses the LLM configured via LLM_KEY in your .env file.
-
- Example 1 - Using environment variables (recommended):
- ```python
- from skyvern import Skyvern
- from skyvern.forge.sdk.api.llm.models import LLMConfig
-
- # Assumes OPENAI_API_KEY is set in your environment
- llm_config = LLMConfig(
- model_name="gpt-4o",
- required_env_vars=["OPENAI_API_KEY"],
- supports_vision=True,
- add_assistant_prefix=False,
- )
- skyvern = Skyvern(llm_config=llm_config)
- ```
-
- Example 2 - Explicitly providing credentials:
- ```python
- from skyvern import Skyvern
- from skyvern.forge.sdk.api.llm.models import LLMConfig, LiteLLMParams
-
- llm_config = LLMConfig(
- model_name="gpt-4o",
- required_env_vars=[], # No env vars required
- supports_vision=True,
- add_assistant_prefix=False,
- litellm_params=LiteLLMParams(
- api_base="https://api.openai.com/v1",
- api_key="sk-...", # Your API key
- ),
- )
- skyvern = Skyvern(llm_config=llm_config)
- ```
- settings: Optional dictionary of Skyvern settings to override.
- These override the corresponding settings from your .env file.
- Example: {"MAX_STEPS_PER_RUN": 100, "BROWSER_TYPE": "chromium-headful"}
+ openai_api_key: Optional OpenAI API key override for LLM operations.
+ If not provided, the one from the .env file will be used.
"""
...
@@ -163,13 +122,12 @@ class Skyvern(AsyncSkyvern):
self,
*,
environment: SkyvernEnvironment | None = None,
+ openai_api_key: str | None = None,
base_url: str | None = None,
api_key: str | None = None,
timeout: float | None = None,
follow_redirects: bool | None = True,
httpx_client: httpx.AsyncClient | None = None,
- llm_config: LLMRouterConfig | LLMConfig | None = None,
- settings: dict[str, Any] | None = None,
):
if environment is None:
if httpx_client is not None:
@@ -189,8 +147,7 @@ class Skyvern(AsyncSkyvern):
timeout=timeout,
follow_redirects=follow_redirects,
httpx_client=create_embedded_server(
- llm_config=llm_config,
- settings_overrides=settings,
+ openai_api_key=openai_api_key,
),
)
else:
diff --git a/skyvern/webeye/actions/handler.py b/skyvern/webeye/actions/handler.py
index 6d8f6822f..8f18edb58 100644
--- a/skyvern/webeye/actions/handler.py
+++ b/skyvern/webeye/actions/handler.py
@@ -2338,7 +2338,8 @@ async def chain_click(
locator=locator,
)
if bound_locator := await skyvern_element.find_bound_label_by_attr_id():
- await bound_locator.click(timeout=timeout)
+ # click on (0, 0) to avoid playwright clicking on the wrong element by accident
+ await bound_locator.click(timeout=timeout, position={"x": 0, "y": 0})
action_results.append(ActionSuccess())
return action_results
except Exception as e:
@@ -2354,7 +2355,8 @@ async def chain_click(
locator=locator,
)
if bound_locator := await skyvern_element.find_bound_label_by_direct_parent():
- await bound_locator.click(timeout=timeout)
+ # click on (0, 0) to avoid playwright clicking on the wrong element by accident
+ await bound_locator.click(timeout=timeout, position={"x": 0, "y": 0})
action_results.append(ActionSuccess())
return action_results
except Exception as e: