backend changes extracted from codex/jon/SKY-5016 (#2508)

skyvern/config.py

@@ -264,6 +264,28 @@ class Settings(BaseSettings):
     SKYVERN_BASE_URL: str = "https://api.skyvern.com"
     SKYVERN_API_KEY: str = "PLACEHOLDER"
 
+    def get_model_name_to_llm_key(self) -> dict[str, str]:
+        """
+        Keys are model names available to blocks in the frontend. These map to key names
+        in LLMConfigRegistry._configs.
+        """
+
+        if self.is_cloud_environment():
+            return {
+                "Gemini 2.5": "GEMINI_2.5_PRO_PREVIEW",
+                "Gemini 2.5 Flash": "VERTEX_GEMINI_2.5_FLASH_PREVIEW_05_20",
+                "GPT 4.1": "OPENAI_GPT4_1",
+                "GPT o3-mini": "OPENAI_O3_MINI",
+            }
+        else:
+            # TODO: apparently the list for OSS is to be much larger
+            return {
+                "Gemini 2.5": "GEMINI_2.5_PRO_PREVIEW",
+                "Gemini 2.5 Flash": "VERTEX_GEMINI_2.5_FLASH_PREVIEW_05_20",
+                "GPT 4.1": "OPENAI_GPT4_1",
+                "GPT o3-mini": "OPENAI_O3_MINI",
+            }
+
     def is_cloud_environment(self) -> bool:
         """
         :return: True if env is not local, else False

skyvern/forge/agent.py

@@ -381,13 +381,15 @@ class ForgeAgent:
             if page := await browser_state.get_working_page():
                 await self.register_async_operations(organization, task, page)
 
-            llm_caller = LLMCallerManager.get_llm_caller(task.task_id)
+            if not llm_caller:
-            if engine == RunEngine.anthropic_cua and not llm_caller:
-                # llm_caller = LLMCaller(llm_key="BEDROCK_ANTHROPIC_CLAUDE3.5_SONNET_INFERENCE_PROFILE")
                 llm_caller = LLMCallerManager.get_llm_caller(task.task_id)
-                if not llm_caller:
+                if engine == RunEngine.anthropic_cua and not llm_caller:
-                    llm_caller = LLMCaller(llm_key=settings.ANTHROPIC_CUA_LLM_KEY, screenshot_scaling_enabled=True)
+                    # llm_caller = LLMCaller(llm_key="BEDROCK_ANTHROPIC_CLAUDE3.5_SONNET_INFERENCE_PROFILE")
-                    LLMCallerManager.set_llm_caller(task.task_id, llm_caller)
+                    llm_caller = LLMCallerManager.get_llm_caller(task.task_id)
+                    if not llm_caller:
+                        llm_caller = LLMCaller(llm_key=settings.ANTHROPIC_CUA_LLM_KEY, screenshot_scaling_enabled=True)
+                        LLMCallerManager.set_llm_caller(task.task_id, llm_caller)
+
             step, detailed_output = await self.agent_step(
                 task,
                 step,

skyvern/forge/sdk/api/llm/config_registry.py

@@ -50,6 +50,10 @@ class LLMConfigRegistry:
 
         return cls._configs[llm_key]
 
+    @classmethod
+    def get_model_names(cls) -> list[str]:
+        return list(cls._configs.keys())
+
 
 if settings.ENABLE_OPENAI:
     LLMConfigRegistry.register_config(

skyvern/forge/sdk/db/client.py

@@ -1181,6 +1181,7 @@ class AgentDB:
         totp_verification_url: str | None = None,
         totp_identifier: str | None = None,
         persist_browser_session: bool = False,
+        model: dict[str, Any] | None = None,
         workflow_permanent_id: str | None = None,
         version: int | None = None,
         is_saved_task: bool = False,
@@ -1197,6 +1198,7 @@ class AgentDB:
                 totp_verification_url=totp_verification_url,
                 totp_identifier=totp_identifier,
                 persist_browser_session=persist_browser_session,
+                model=model,
                 is_saved_task=is_saved_task,
                 status=status,
             )

skyvern/forge/sdk/db/models.py

@@ -216,6 +216,7 @@ class WorkflowModel(Base):
     totp_verification_url = Column(String)
     totp_identifier = Column(String)
     persist_browser_session = Column(Boolean, default=False, nullable=False)
+    model = Column(JSON, nullable=True)
     status = Column(String, nullable=False, default="published")
 
     created_at = Column(DateTime, default=datetime.datetime.utcnow, nullable=False)

skyvern/forge/sdk/db/utils.py

@@ -185,6 +185,7 @@ def convert_to_workflow(workflow_model: WorkflowModel, debug_enabled: bool = Fal
         totp_verification_url=workflow_model.totp_verification_url,
         totp_identifier=workflow_model.totp_identifier,
         persist_browser_session=workflow_model.persist_browser_session,
+        model=workflow_model.model,
         proxy_location=(ProxyLocation(workflow_model.proxy_location) if workflow_model.proxy_location else None),
         version=workflow_model.version,
         is_saved_task=workflow_model.is_saved_task,

skyvern/forge/sdk/routes/agent_protocol.py

@@ -48,6 +48,7 @@ from skyvern.forge.sdk.schemas.task_generations import GenerateTaskRequest, Task
 from skyvern.forge.sdk.schemas.task_v2 import TaskV2Request
 from skyvern.forge.sdk.schemas.tasks import (
     CreateTaskResponse,
+    ModelsResponse,
     OrderBy,
     SortDirection,
     Task,
@@ -794,6 +795,24 @@ async def heartbeat() -> Response:
     return Response(content="Server is running.", status_code=200, headers={"X-Skyvern-API-Version": __version__})
 
 
+@legacy_base_router.get(
+    "/models",
+    tags=["agent"],
+    openapi_extra={
+        "x-fern-sdk-group-name": "agent",
+    },
+)
+@legacy_base_router.get("/models/", include_in_schema=False)
+async def models() -> ModelsResponse:
+    """
+    Get a list of available models.
+    """
+    mapping = settings.get_model_name_to_llm_key()
+    models = list(mapping.keys())
+
+    return ModelsResponse(models=models)
+
+
 @legacy_base_router.post(
     "/tasks",
     tags=["agent"],

skyvern/forge/sdk/routes/code_samples.py

@@ -36,6 +36,8 @@ proxy_location: RESIDENTIAL
 webhook_callback_url: https://example.com/webhook
 totp_verification_url: https://example.com/totp
 persist_browser_session: false
+model:
+  model: gpt-3.5-turbo
 workflow_definition:
   parameters:
     - key: website_url
@@ -119,6 +121,7 @@ workflow_definition = {
     "webhook_callback_url": "https://example.com/webhook",
     "totp_verification_url": "https://example.com/totp",
     "totp_identifier": "4155555555",
+    "model": {"model": "gpt-3.5-turbo"},
     "workflow_definition": {
         "parameters": [
             {
@@ -201,6 +204,7 @@ proxy_location: RESIDENTIAL
 webhook_callback_url: https://example.com/webhook
 totp_verification_url: https://example.com/totp
 persist_browser_session: false
+model: {model: gpt-3.5-turbo}
 workflow_definition:
   parameters:
     - key: website_url
@@ -283,6 +287,7 @@ updated_workflow_definition = {
     "webhook_callback_url": "https://example.com/webhook",
     "totp_verification_url": "https://example.com/totp",
     "totp_identifier": "4155555555",
+    "model": {"model": "gpt-3.5-turbo"},
     "workflow_definition": {
         "parameters": [
             {

skyvern/forge/sdk/schemas/tasks.py

@@ -349,3 +349,7 @@ class OrderBy(StrEnum):
 class SortDirection(StrEnum):
     asc = "asc"
     desc = "desc"
+
+
+class ModelsResponse(BaseModel):
+    models: list[str]

skyvern/forge/sdk/workflow/models/block.py

@@ -46,7 +46,7 @@ from skyvern.forge.sdk.api.files import (
     download_from_s3,
     get_path_for_workflow_download_directory,
 )
-from skyvern.forge.sdk.api.llm.api_handler_factory import LLMAPIHandlerFactory
+from skyvern.forge.sdk.api.llm.api_handler_factory import LLMAPIHandlerFactory, LLMCaller
 from skyvern.forge.sdk.artifact.models import ArtifactType
 from skyvern.forge.sdk.core import skyvern_context
 from skyvern.forge.sdk.db.enums import TaskType
@@ -126,6 +126,7 @@ class Block(BaseModel, abc.ABC):
     block_type: BlockType
     output_parameter: OutputParameter
     continue_on_failure: bool = False
+    model: dict[str, Any] | None = None
 
     async def record_output_parameter_value(
         self,
@@ -618,6 +619,9 @@ class BaseTaskBlock(Block):
             try:
                 current_context = skyvern_context.ensure_context()
                 current_context.task_id = task.task_id
+                llm_key = workflow.determine_llm_key(block=self)
+                llm_caller = None if not llm_key else LLMCaller(llm_key=llm_key)
+
                 await app.agent.execute_step(
                     organization=organization,
                     task=task,
@@ -627,6 +631,7 @@ class BaseTaskBlock(Block):
                     close_browser_on_completion=browser_session_id is None,
                     complete_verification=self.complete_verification,
                     engine=self.engine,
+                    llm_caller=llm_caller,
                 )
             except Exception as e:
                 # Make sure the task is marked as failed in the database before raising the exception

skyvern/forge/sdk/workflow/models/workflow.py

@@ -5,10 +5,11 @@ from typing import Any, List
 from pydantic import BaseModel, field_validator
 from typing_extensions import deprecated
 
+from skyvern.config import settings
 from skyvern.forge.sdk.schemas.files import FileInfo
 from skyvern.forge.sdk.schemas.task_v2 import TaskV2
 from skyvern.forge.sdk.workflow.exceptions import WorkflowDefinitionHasDuplicateBlockLabels
-from skyvern.forge.sdk.workflow.models.block import BlockTypeVar
+from skyvern.forge.sdk.workflow.models.block import Block, BlockTypeVar
 from skyvern.forge.sdk.workflow.models.parameter import PARAMETER_TYPE
 from skyvern.schemas.runs import ProxyLocation
 from skyvern.utils.url_validators import validate_url
@@ -74,12 +75,41 @@ class Workflow(BaseModel):
     totp_verification_url: str | None = None
     totp_identifier: str | None = None
     persist_browser_session: bool = False
+    model: dict[str, Any] | None = None
     status: WorkflowStatus = WorkflowStatus.published
 
     created_at: datetime
     modified_at: datetime
     deleted_at: datetime | None = None
 
+    def determine_llm_key(self, *, block: Block | None = None) -> str | None:
+        """
+        Determine the LLM key override to use for a block, if it has one.
+
+        It has one if:
+            - it defines one, or
+            - the workflow it is a part of (if applicable) defines one
+        """
+
+        mapping = settings.get_model_name_to_llm_key()
+
+        if block:
+            model_name = (block.model or {}).get("model")
+
+            if model_name:
+                llm_key = mapping.get(model_name)
+                if llm_key:
+                    return llm_key
+
+        workflow_model_name = (self.model or {}).get("model")
+
+        if workflow_model_name:
+            llm_key = mapping.get(workflow_model_name)
+            if llm_key:
+                return llm_key
+
+        return None
+
 
 class WorkflowRunStatus(StrEnum):
     created = "created"

skyvern/forge/sdk/workflow/models/yaml.py

@@ -117,6 +117,7 @@ class BlockYAML(BaseModel, abc.ABC):
     block_type: BlockType
     label: str
     continue_on_failure: bool = False
+    model: dict[str, Any] | None = None
 
 
 class TaskBlockYAML(BlockYAML):
@@ -413,6 +414,7 @@ class WorkflowCreateYAMLRequest(BaseModel):
     totp_verification_url: str | None = None
     totp_identifier: str | None = None
     persist_browser_session: bool = False
+    model: dict[str, Any] | None = None
     workflow_definition: WorkflowDefinitionYAML
     is_saved_task: bool = False
     status: WorkflowStatus = WorkflowStatus.published

skyvern/forge/sdk/workflow/service.py

@@ -180,7 +180,7 @@ class WorkflowService:
                 else:
                     raise MissingValueForParameter(
                         parameter_key=workflow_parameter.key,
-                        workflow_id=workflow.workflow_permanent_id,
+                        workflow_id=workflow.workflow_id,
                         workflow_run_id=workflow_run.workflow_run_id,
                     )
         except Exception as e:
@@ -336,6 +336,7 @@ class WorkflowService:
                     block_idx=block_idx,
                     block_type_var=block.block_type,
                     block_label=block.label,
+                    model=block.model,
                 )
                 block_result = await block.execute_safe(
                     workflow_run_id=workflow_run_id,
@@ -550,6 +551,7 @@ class WorkflowService:
         totp_verification_url: str | None = None,
         totp_identifier: str | None = None,
         persist_browser_session: bool = False,
+        model: dict[str, Any] | None = None,
         workflow_permanent_id: str | None = None,
         version: int | None = None,
         is_saved_task: bool = False,
@@ -565,6 +567,7 @@ class WorkflowService:
             totp_verification_url=totp_verification_url,
             totp_identifier=totp_identifier,
             persist_browser_session=persist_browser_session,
+            model=model,
             workflow_permanent_id=workflow_permanent_id,
             version=version,
             is_saved_task=is_saved_task,
@@ -592,6 +595,7 @@ class WorkflowService:
         )
         if not workflow:
             raise WorkflowNotFound(workflow_permanent_id=workflow_permanent_id, version=version)
+
         return workflow
 
     async def get_workflows_by_permanent_ids(
@@ -1403,6 +1407,7 @@ class WorkflowService:
                     totp_verification_url=request.totp_verification_url,
                     totp_identifier=request.totp_identifier,
                     persist_browser_session=request.persist_browser_session,
+                    model=request.model,
                     workflow_permanent_id=workflow_permanent_id,
                     version=existing_version + 1,
                     is_saved_task=request.is_saved_task,
@@ -1419,6 +1424,7 @@ class WorkflowService:
                     totp_verification_url=request.totp_verification_url,
                     totp_identifier=request.totp_identifier,
                     persist_browser_session=request.persist_browser_session,
+                    model=request.model,
                     is_saved_task=request.is_saved_task,
                     status=request.status,
                 )
@@ -1657,6 +1663,7 @@ class WorkflowService:
                 error_code_mapping=block_yaml.error_code_mapping,
                 max_steps_per_run=block_yaml.max_steps_per_run,
                 max_retries=block_yaml.max_retries,
+                model=block_yaml.model,
                 complete_on_download=block_yaml.complete_on_download,
                 download_suffix=block_yaml.download_suffix,
                 continue_on_failure=block_yaml.continue_on_failure,
@@ -1724,6 +1731,7 @@ class WorkflowService:
                 json_schema=block_yaml.json_schema,
                 output_parameter=output_parameter,
                 continue_on_failure=block_yaml.continue_on_failure,
+                model=block_yaml.model,
             )
         elif block_yaml.block_type == BlockType.DOWNLOAD_TO_S3:
             return DownloadToS3Block(
@@ -1781,6 +1789,7 @@ class WorkflowService:
                 file_url=block_yaml.file_url,
                 json_schema=block_yaml.json_schema,
                 continue_on_failure=block_yaml.continue_on_failure,
+                model=block_yaml.model,
             )
         elif block_yaml.block_type == BlockType.VALIDATION:
             validation_block_parameters = (
@@ -1803,6 +1812,7 @@ class WorkflowService:
                 continue_on_failure=block_yaml.continue_on_failure,
                 # only need one step for validation block
                 max_steps_per_run=1,
+                model=block_yaml.model,
             )
 
         elif block_yaml.block_type == BlockType.ACTION:
@@ -1826,6 +1836,7 @@ class WorkflowService:
                 navigation_goal=block_yaml.navigation_goal,
                 error_code_mapping=block_yaml.error_code_mapping,
                 max_retries=block_yaml.max_retries,
+                model=block_yaml.model,
                 complete_on_download=block_yaml.complete_on_download,
                 download_suffix=block_yaml.download_suffix,
                 continue_on_failure=block_yaml.continue_on_failure,
@@ -1854,6 +1865,7 @@ class WorkflowService:
                 error_code_mapping=block_yaml.error_code_mapping,
                 max_steps_per_run=block_yaml.max_steps_per_run,
                 max_retries=block_yaml.max_retries,
+                model=block_yaml.model,
                 complete_on_download=block_yaml.complete_on_download,
                 download_suffix=block_yaml.download_suffix,
                 continue_on_failure=block_yaml.continue_on_failure,
@@ -1883,6 +1895,7 @@ class WorkflowService:
                 data_schema=block_yaml.data_schema,
                 max_steps_per_run=block_yaml.max_steps_per_run,
                 max_retries=block_yaml.max_retries,
+                model=block_yaml.model,
                 continue_on_failure=block_yaml.continue_on_failure,
                 cache_actions=block_yaml.cache_actions,
                 complete_verification=False,
@@ -1905,6 +1918,7 @@ class WorkflowService:
                 error_code_mapping=block_yaml.error_code_mapping,
                 max_steps_per_run=block_yaml.max_steps_per_run,
                 max_retries=block_yaml.max_retries,
+                model=block_yaml.model,
                 continue_on_failure=block_yaml.continue_on_failure,
                 totp_verification_url=block_yaml.totp_verification_url,
                 totp_identifier=block_yaml.totp_identifier,
@@ -1942,6 +1956,7 @@ class WorkflowService:
                 error_code_mapping=block_yaml.error_code_mapping,
                 max_steps_per_run=block_yaml.max_steps_per_run,
                 max_retries=block_yaml.max_retries,
+                model=block_yaml.model,
                 download_suffix=block_yaml.download_suffix,
                 continue_on_failure=block_yaml.continue_on_failure,
                 totp_verification_url=block_yaml.totp_verification_url,
@@ -1959,6 +1974,7 @@ class WorkflowService:
                 totp_identifier=block_yaml.totp_identifier,
                 max_iterations=block_yaml.max_iterations,
                 max_steps=block_yaml.max_steps,
+                model=block_yaml.model,
                 output_parameter=output_parameter,
             )
         elif block_yaml.block_type == BlockType.GOTO_URL: