feat: added Grok as a provider and related functionality

.github/workflows/publish.yaml

@@ -43,6 +43,7 @@ jobs:
           OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
           ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
           OPENROUTER_API_KEY: ${{ secrets.OPENROUTER_API_KEY }}
+          GROK_API_KEY: ${{ secrets.GROK_API_KEY }}
 
   publish:
     runs-on: ubuntu-latest

.vscode/settings.json

@@ -7,5 +7,8 @@
     "editor.tabSize": 4,
     "editor.insertSpaces": true,
     "files.trimTrailingWhitespace": true,
-    "files.insertFinalNewline": true
+    "files.insertFinalNewline": true,
+    "python.linting.flake8Args": [
+        "--config=pyproject.toml"
+    ]
 }

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "emp-agents"
-version = "0.3.0"
+version = "0.3.1"
 description = "A Python library for building low-code, capable and extensible autonomous agent systems."
 readme = "README.md"
 requires-python = ">=3.10"
@@ -76,3 +76,12 @@ profile = "black"
 line_length = 88
 known_third_party = ["pytest", "black", "isort"]
 default_section = "THIRDPARTY"
+
+[tool.poetry.group.dev.dependencies]
+flake8-pyproject = "^1.2.3"
+pytest = "^7.0.0"
+pytest-asyncio = "^0.23.0"
+
+[tool.flake8]
+ignore = ["W503"]
+max-line-length = 160

src/emp_agents/__init__.py

@@ -10,7 +10,7 @@
     SystemMessage,
     UserMessage,
 )
-from .providers import AnthropicProvider, DeepSeekProvider, OpenAIProvider
+from .providers import AnthropicProvider, DeepSeekProvider, GrokProvider, OpenAIProvider
 from .types import Role
 
 __all__ = [
@@ -19,6 +19,7 @@
     "AssistantMessage",
     "DeepSeekProvider",
     "GenericTool",
+    "GrokProvider",
     "Message",
     "Middleware",
     "OpenAIProvider",

src/emp_agents/models/shared/tools.py

@@ -136,6 +136,30 @@ def to_anthropic(self):
             ),
         )
 
+    def to_grok(self):
+        from emp_agents.providers.openai.tool import (
+            Function,
+            Parameters,
+            Property,
+            Tool,
+        )
+
+        return Tool(
+            type="function",
+            function=Function(
+                description=self.description,
+                name=self.name,
+                parameters=Parameters(
+                    properties={
+                        key: Property(**param.model_dump(exclude_none=True))
+                        for key, param in self.parameters.items()
+                    },
+                    required=self.required,
+                ),
+            ),
+            strict=True,
+        )
+
     def __repr__(self):
         description = self.description.strip().replace("\n", " ")[:100]
         if len(description) >= 50:

src/emp_agents/providers/__init__.py

@@ -1,15 +1,20 @@
 from .anthropic import AnthropicModelType, AnthropicProvider
 from .deepseek import DeepSeekModelType, DeepSeekProvider
+from .grok import GrokModelType, GrokProvider
 from .openai import OpenAIModelType, OpenAIProvider
 from .openrouter import OpenRouterModelType, OpenRouterProvider
+from .standard_request import StandardRequest
 
 __all__ = [
     "AnthropicProvider",
     "DeepSeekProvider",
+    "GrokProvider",
     "OpenAIProvider",
     "AnthropicModelType",
     "DeepSeekModelType",
+    "GrokModelType",
     "OpenAIModelType",
     "OpenRouterProvider",
     "OpenRouterModelType",
+    "StandardRequest",
 ]

src/emp_agents/providers/anthropic/__init__.py

@@ -59,7 +59,7 @@ def _from_request(self, request: Request):
             message for message in request.messages if message.role == Role.system
         ]
         result["system"] = result.get("system", "") + str(
-            "\n".join([m.content for m in system_messages])
+            "\n".join([m.content for m in system_messages if m.content is not None])
         )
         result["messages"] = [m.model_dump(exclude_none=True) for m in messages]
 

src/emp_agents/providers/anthropic/response.py

@@ -3,7 +3,8 @@
 
 from pydantic import BaseModel
 
-from emp_agents.models.shared import AssistantMessage, ToolCall
+from emp_agents.models.provider import ResponseT
+from emp_agents.models.shared import AssistantMessage, Message, ToolCall
 from emp_agents.types import Role
 
 from .types import AnthropicModelType as ModelType
@@ -32,18 +33,18 @@ def to_message(self) -> AssistantMessage:
 
 
 class StopReason(StrEnum):
-    tool_use: str = "tool_use"
-    end_turn: str = "end_turn"
-    max_tokens: str = "max_tokens"
-    stop_sequence: str = "stop_sequence"
+    tool_use = "tool_use"
+    end_turn = "end_turn"
+    max_tokens = "max_tokens"
+    stop_sequence = "stop_sequence"
 
 
 class Usage(BaseModel):
     input_tokens: int
     output_tokens: int
 
 
-class Response(BaseModel):
+class Response(ResponseT):
     id: str
     content: list[Content]
     model: ModelType | str
@@ -62,5 +63,5 @@ def tool_calls(self):
         return [content for content in self.content if content.type == "tool_use"]
 
     @property
-    def messages(self) -> list[AssistantMessage]:
+    def messages(self) -> list[Message]:
         return [content.to_message() for content in self.content]

src/emp_agents/providers/grok/__init__.py

@@ -0,0 +1,28 @@
+import os
+from typing import ClassVar
+
+from pydantic import Field
+
+from emp_agents.providers.openai import OpenAIProviderBase
+from emp_agents.providers.openai.response import Response
+
+from .types import GrokModelType
+
+
+class GrokProvider(OpenAIProviderBase[GrokModelType]):
+    """
+    Provider for Grok API, which follows the OpenAI API format.
+    This inherits from OpenAIProvider since the APIs are compatible.
+    """
+
+    URL: ClassVar[str] = "https://api.x.ai/v1/chat/completions"
+
+    api_key: str = Field(default_factory=lambda: os.environ["GROK_API_KEY"])
+    default_model: GrokModelType = Field(default=GrokModelType.grok_2)
+
+
+__all__ = [
+    "GrokModelType",
+    "GrokProvider",
+    "Response",
+]

src/emp_agents/providers/grok/request.py

@@ -0,0 +1,16 @@
+from typing import Optional
+
+from pydantic import Field
+
+from emp_agents.models.shared.tools import GenericTool
+from emp_agents.providers.standard_request import StandardRequest
+
+from .types import GrokModelType
+
+
+class Request(StandardRequest[GrokModelType]):
+    """
+    Request model for Grok API, which follows the OpenAI API format.
+    """
+
+    tools: Optional[list[GenericTool]] = Field(default=None)

src/emp_agents/providers/grok/response.py

@@ -0,0 +1,7 @@
+from emp_agents.providers.openai.response import Response as OpenAIResponse
+
+from .types import GrokModelType
+
+
+class Response(OpenAIResponse):
+    model: GrokModelType | str

src/emp_agents/providers/grok/types.py

@@ -0,0 +1,8 @@
+from enum import StrEnum
+
+
+class GrokModelType(StrEnum):
+    grok_1 = "grok-1"
+    grok_1_5 = "grok-1.5"
+    grok_2 = "grok-2"
+    grok_3 = "grok-3-latest"

src/emp_agents/providers/openai/__init__.py

@@ -1,22 +1,25 @@
 import os
-from typing import Any, ClassVar
+from typing import Any, ClassVar, Generic, TypeVar
 
 import httpx
 from pydantic import Field
 
 from emp_agents.models import GenericTool, Provider, Request, SystemMessage
+from emp_agents.models.shared.message import Message
 
-from .request import Message, Tool
+from .request import Tool
 from .response import Response
 from .tool import Function, Parameters, Property
 from .types import Classification, OpenAIModelType
 
+ModelType = TypeVar("ModelType", bound=str)
 
-class OpenAIProvider(Provider[Response]):
+
+class OpenAIProviderBase(Provider[Response], Generic[ModelType]):
     URL: ClassVar[str] = "https://api.openai.com/v1/chat/completions"
 
     api_key: str = Field(default_factory=lambda: os.environ["OPENAI_API_KEY"])
-    default_model: OpenAIModelType = Field(default=OpenAIModelType.gpt4o_mini)
+    default_model: ModelType
 
     @property
     def headers(self):
@@ -130,6 +133,10 @@ def to_tool_call(self, tool: GenericTool):
         )
 
 
+class OpenAIProvider(OpenAIProviderBase[OpenAIModelType]):
+    default_model: OpenAIModelType = Field(default=OpenAIModelType.gpt4o_mini)
+
+
 __all__ = [
     "Classification",
     "Message",

src/emp_agents/providers/openai/request.py

@@ -1,25 +1,19 @@
-from typing import Annotated, Literal, Optional
+from typing import Annotated, Optional
 
-from pydantic import BaseModel, ConfigDict, Field, PlainSerializer
+from pydantic import Field, PlainSerializer
 
-from emp_agents.models.shared import Message
 from emp_agents.models.shared.tools import GenericTool
+from emp_agents.providers.standard_request import StandardRequest
 
 from .tool import Tool
 from .types import OpenAIModelType
 
 
-class Request(BaseModel):
+class Request(StandardRequest[OpenAIModelType]):
     """
     https://platform.openai.com/docs/api-reference/chat/create
     """
 
-    model_config = ConfigDict(populate_by_name=True)
-
-    model: OpenAIModelType
-    max_tokens: Optional[int] = Field(default=None)
-    temperature: Optional[float] = Field(default=None, ge=0, le=2.0)
-    tool_choice: Literal["none", "required", "auto", None] = Field(default=None)
     tools: Annotated[
         Optional[list[GenericTool]],
         PlainSerializer(
@@ -31,18 +25,3 @@ class Request(BaseModel):
             return_type=Optional[list[Tool]],
         ),
     ] = Field(default=None)
-
-    system: str | None = None
-    messages: list[Message] | None = None  # Anthropic Field
-
-    frequency_penalty: Optional[float] = Field(default=None, ge=-2.0, le=2.0)  # openai
-    presence_penalty: Optional[float] = Field(default=None, ge=-2.0, le=2.0)  # openai
-    num_responses: Optional[int] = Field(
-        default=None, serialization_alias="n"
-    )  # openai
-    top_p: Optional[int] = Field(default=None)  # openai
-
-    def model_dump(self, *, exclude_none=True, by_alias=True, **kwargs):
-        return super().model_dump(
-            exclude_none=exclude_none, by_alias=by_alias, **kwargs
-        )

src/emp_agents/providers/standard_request.py

@@ -0,0 +1,34 @@
+from typing import Annotated, Generic, Literal, Optional, TypeVar
+
+from pydantic import BaseModel, ConfigDict, Field, PlainSerializer
+
+from emp_agents.models.shared import Message
+from emp_agents.models.shared.tools import GenericTool
+
+ModelType = TypeVar("ModelType")
+
+
+class StandardRequest(BaseModel, Generic[ModelType]):
+    """
+    Generic request model that can be used by all platforms that mimic the OpenAI format.
+    """
+
+    model_config = ConfigDict(populate_by_name=True)
+
+    model: ModelType
+    max_tokens: Optional[int] = Field(default=None)
+    temperature: Optional[float] = Field(default=None, ge=0, le=2.0)
+    tool_choice: Literal["none", "required", "auto", None] = Field(default=None)
+
+    system: str | None = None
+    messages: list[Message] | None = None
+
+    frequency_penalty: Optional[float] = Field(default=None, ge=-2.0, le=2.0)
+    presence_penalty: Optional[float] = Field(default=None, ge=-2.0, le=2.0)
+    num_responses: Optional[int] = Field(default=None, serialization_alias="n")
+    top_p: Optional[int] = Field(default=None)
+
+    def model_dump(self, *, exclude_none=True, by_alias=True, **kwargs):
+        return super().model_dump(
+            exclude_none=exclude_none, by_alias=by_alias, **kwargs
+        )