feat(integration-service): Add integrations service (#520)

Introduces a new integration service with FastAPI, Docker setup, and support for multiple integrations like DALL-E and Wikipedia.

- Integration Service:
  + Adds integrations-service with Docker setup in docker-compose.yml and Dockerfile.
  + FastAPI application in web.py with routers for execution and integration management.
- Models:
  + Defines models for DalleImageGenerator, DuckDuckGoSearch, HackerNews, Weather, and Wikipedia in models directory.
  + IntegrationExecutionRequest and IntegrationExecutionResponse for handling execution requests.
- Utilities:
  + Implements execute_integration in execute_integration.py to handle integration execution logic.
  + Integration utilities for DALL-E, DuckDuckGo, Hacker News, Weather, and Wikipedia in utils/integrations.
- Configuration:
  + Adds pyproject.toml for dependency management with Poetry.
  + Adds pytype.toml for type checking configuration.

docker-compose.yml

@@ -11,6 +11,7 @@ include:
   - ./agents-api/docker-compose.yml
   - ./scheduler/docker-compose.yml
   - ./llm-proxy/docker-compose.yml
+  - ./integrations-service/docker-compose.yml
 
   # TODO: Enable after testing
   # - ./monitoring/docker-compose.yml

integrations-service/Dockerfile

@@ -0,0 +1,19 @@
+FROM python:3.11-slim
+
+WORKDIR /app
+
+# Install Poetry
+RUN pip install poetry
+
+# Copy only requirements to cache them in docker layer
+COPY pyproject.toml poetry.lock* /app/
+
+# Project initialization:
+RUN poetry config virtualenvs.create false \
+    && poetry install --no-interaction --no-ansi
+
+# Copy project
+COPY . ./
+
+# Run the application
+CMD ["python", "-m", "integrations.web", "--host", "0.0.0.0", "--port", "8000"]

integrations-service/docker-compose.yml

@@ -0,0 +1,26 @@
+name: julep-integrations
+
+# Shared environment variables
+x--shared-environment: &shared-environment
+  OPENAI_API_KEY: ${OPENAI_API_KEY}
+
+services:
+  integrations:
+    environment:
+      <<: *shared-environment
+
+    build: .
+    ports:
+      - "8000:8000"
+
+    develop:
+      watch:
+        - action: sync+restart
+          path: ./
+          target: /app/
+          ignore:
+            - ./**/*.pyc
+        - action: rebuild
+          path: poetry.lock
+        - action: rebuild
+          path: Dockerfile

integrations-service/integrations/models/__init__.py

@@ -0,0 +1,17 @@
+from .dalle_image_generator import (
+    DalleImageGeneratorArguments,
+    DalleImageGeneratorSetup,
+)
+from .duckduckgo_search import DuckDuckGoSearchExecutionArguments
+from .hacker_news import HackerNewsExecutionArguments
+
+# TODO: Move these models somewhere else
+from .models import (
+    ExecuteIntegrationArguments,
+    ExecuteIntegrationSetup,
+    IntegrationDef,
+    IntegrationExecutionRequest,
+    IntegrationExecutionResponse,
+)
+from .weather import WeatherExecutionArguments, WeatherExecutionSetup
+from .wikipedia import WikipediaExecutionArguments

integrations-service/integrations/models/dalle_image_generator.py

@@ -0,0 +1,9 @@
+from pydantic import BaseModel, Field
+
+
+class DalleImageGeneratorSetup(BaseModel):
+    api_key: str = Field(str, description="The API key for DALL-E")
+
+
+class DalleImageGeneratorArguments(BaseModel):
+    prompt: str = Field(str, description="The image generation prompt")

integrations-service/integrations/models/duckduckgo_search.py

@@ -0,0 +1,5 @@
+from pydantic import BaseModel, Field
+
+
+class DuckDuckGoSearchExecutionArguments(BaseModel):
+    query: str = Field(..., description="The search query string")

integrations-service/integrations/models/hacker_news.py

@@ -0,0 +1,5 @@
+from pydantic import BaseModel, Field
+
+
+class HackerNewsExecutionArguments(BaseModel):
+    url: str = Field(..., description="The URL of the Hacker News thread to fetch")

integrations-service/integrations/models/models.py

@@ -0,0 +1,81 @@
+from typing import Literal, Union
+
+from pydantic import BaseModel
+
+from .dalle_image_generator import (
+    DalleImageGeneratorArguments,
+    DalleImageGeneratorSetup,
+)
+from .duckduckgo_search import DuckDuckGoSearchExecutionArguments
+from .hacker_news import HackerNewsExecutionArguments
+from .weather import WeatherExecutionArguments, WeatherExecutionSetup
+from .wikipedia import WikipediaExecutionArguments
+
+ExecuteIntegrationArguments = Union[
+    WikipediaExecutionArguments,
+    DuckDuckGoSearchExecutionArguments,
+    DalleImageGeneratorArguments,
+    WeatherExecutionArguments,
+    HackerNewsExecutionArguments,
+]
+
+ExecuteIntegrationSetup = Union[
+    DalleImageGeneratorSetup,
+    WeatherExecutionSetup,
+]
+
+
+class IntegrationExecutionRequest(BaseModel):
+    setup: ExecuteIntegrationSetup | None = None
+    """
+    The setup parameters the integration accepts (such as API keys)
+    """
+    arguments: ExecuteIntegrationArguments
+    """
+    The arguments to pass to the integration
+    """
+
+
+class IntegrationExecutionResponse(BaseModel):
+    result: str
+    """
+    The result of the integration execution
+    """
+
+
+class IntegrationDef(BaseModel):
+    provider: (
+        Literal[
+            "dummy",
+            "dalle_image_generator",
+            "duckduckgo_search",
+            "hacker_news",
+            "weather",
+            "wikipedia",
+            "twitter",
+            "web_base",
+            "requests",
+            "gmail",
+            "tts_query",
+        ]
+        | None
+    ) = None
+    """
+    The provider of the integration
+    """
+    method: str | None = None
+    """
+    The specific method of the integration to call
+    """
+    description: str | None = None
+    """
+    Optional description of the integration
+    """
+    setup: dict | None = None
+    """
+    The setup parameters the integration accepts
+    """
+    arguments: dict | None = None
+    """
+    The arguments to pre-apply to the integration call
+    """

integrations-service/integrations/models/weather.py

@@ -0,0 +1,13 @@
+from pydantic import BaseModel, Field
+
+
+class WeatherExecutionSetup(BaseModel):
+    openweathermap_api_key: str = Field(
+        ..., description="The location for which to fetch weather data"
+    )
+
+
+class WeatherExecutionArguments(BaseModel):
+    location: str = Field(
+        ..., description="The location for which to fetch weather data"
+    )

integrations-service/integrations/models/wikipedia.py

@@ -0,0 +1,6 @@
+from pydantic import BaseModel, Field
+
+
+class WikipediaExecutionArguments(BaseModel):
+    query: str = Field(..., description="The search query string")
+    load_max_docs: int = Field(2, description="Maximum number of documents to load")

integrations-service/integrations/routers/__init__.py

@@ -0,0 +1,2 @@
+from .execution.router import router as execution_router
+from .integrations.router import router as integrations_router

integrations-service/integrations/routers/execution/__init__.py

@@ -0,0 +1 @@
+from .execute import execute

integrations-service/integrations/routers/execution/execute.py

@@ -0,0 +1,19 @@
+from fastapi import Body, HTTPException, Path
+
+from ...models import IntegrationExecutionRequest, IntegrationExecutionResponse
+from ...utils.execute_integration import execute_integration
+from .router import router
+
+
+@router.post("/execute/{provider}", tags=["execution"])
+async def execute(
+    provider: str = Path(..., description="The integration provider"),
+    request: IntegrationExecutionRequest = Body(
+        ..., description="The integration execution request"
+    ),
+) -> IntegrationExecutionResponse:
+    try:
+        result = await execute_integration(provider, request.setup, request.arguments)
+        return IntegrationExecutionResponse(result=result)
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))

integrations-service/integrations/routers/execution/router.py

@@ -0,0 +1,3 @@
+from fastapi import APIRouter
+
+router: APIRouter = APIRouter()

integrations-service/integrations/routers/integrations/__init__.py

@@ -0,0 +1,2 @@
+from .get_integration_tool import get_integration_tool
+from .get_integrations import get_integrations

integrations-service/integrations/routers/integrations/get_integration_tool.py

@@ -0,0 +1,40 @@
+from typing import Optional
+
+from fastapi import HTTPException
+
+from ...models.models import IntegrationDef
+from .get_integrations import get_integrations
+from .router import router
+
+
+def convert_to_openai_tool(integration: IntegrationDef) -> dict:
+    return {
+        "type": "function",
+        "function": {
+            "name": integration.provider,
+            "description": integration.description,
+            "parameters": {
+                "type": "object",
+                "properties": integration.arguments,
+                "required": [
+                    k
+                    for k, v in integration.arguments.items()
+                    if v.get("required", False)
+                ],
+            },
+        },
+    }
+
+
+@router.get("/integrations/{provider}/tool", tags=["integration_tool"])
+@router.get("/integrations/{provider}/{method}/tool", tags=["integration_tool"])
+async def get_integration_tool(provider: str, method: Optional[str] = None):
+    integrations = await get_integrations()
+
+    for integration in integrations:
+        if integration.provider == provider and (
+            method is None or integration.method == method
+        ):
+            return convert_to_openai_tool(integration)
+
+    raise HTTPException(status_code=404, detail="Integration not found")

integrations-service/integrations/routers/integrations/get_integrations.py

@@ -0,0 +1,82 @@
+import importlib
+import inspect
+import os
+from typing import Any, List
+
+from pydantic import BaseModel
+
+from ...models.models import IntegrationDef
+from ...utils import integrations
+from .router import router
+
+
+def create_integration_def(module: Any) -> IntegrationDef:
+    module_parts = module.__name__.split(".")
+    if len(module_parts) > 4:  # Nested integration
+        provider = module_parts[-2]
+        method = module_parts[-1]
+    else:  # Top-level integration
+        provider = module_parts[-1]
+        method = None
+
+    # Find the first function in the module
+    function_name = next(
+        name
+        for name, obj in inspect.getmembers(module)
+        if inspect.isfunction(obj) and not name.startswith("_")
+    )
+    function = getattr(module, function_name)
+    signature = inspect.signature(function)
+
+    # Get the Pydantic model for the parameters
+    params_model = next(iter(signature.parameters.values())).annotation
+
+    # Check if the params_model is a Pydantic model
+    if issubclass(params_model, BaseModel):
+        arguments = {}
+        for field_name, field in params_model.model_fields.items():
+            field_type = field.annotation
+            arguments[field_name] = {
+                "type": field_type.__name__.lower(),
+                "description": field.description,
+            }
+    else:
+        # Fallback to a dictionary if it's not a Pydantic model
+        arguments = {
+            param.name: {"type": str(param.annotation.__name__).lower()}
+            for param in signature.parameters.values()
+            if param.name != "parameters"
+        }
+
+    return IntegrationDef(
+        provider=provider,
+        method=method,
+        description=function.__doc__.strip() if function.__doc__ else None,
+        arguments=arguments,
+    )
+
+
+@router.get("/integrations", tags=["integrations"])
+async def get_integrations() -> List[IntegrationDef]:
+    integration_defs = []
+    integrations_dir = os.path.dirname(integrations.__file__)
+
+    for item in os.listdir(integrations_dir):
+        item_path = os.path.join(integrations_dir, item)
+
+        if os.path.isdir(item_path):
+            # This is a toolkit
+            for file in os.listdir(item_path):
+                if file.endswith(".py") and not file.startswith("__"):
+                    module = importlib.import_module(
+                        f"...utils.integrations.{item}.{file[:-3]}", package=__package__
+                    )
+                    integration_defs.append(create_integration_def(module))
+        elif item.endswith(".py") and not item.startswith("__"):
+            # This is a single-file tool
+            module = importlib.import_module(
+                f"...utils.integrations.{item[:-3]}", package=__package__
+            )
+            integration_defs.append(create_integration_def(module))
+
+    return integration_defs

integrations-service/integrations/routers/integrations/router.py

@@ -0,0 +1,3 @@
+from fastapi import APIRouter
+
+router: APIRouter = APIRouter()

integrations-service/integrations/utils/execute_integration.py

@@ -0,0 +1,26 @@
+from ..models import ExecuteIntegrationArguments, ExecuteIntegrationSetup
+from .integrations.dalle_image_generator import dalle_image_generator
+from .integrations.duckduckgo_search import duckduckgo_search
+from .integrations.hacker_news import hacker_news
+from .integrations.weather import weather
+from .integrations.wikipedia import wikipedia
+
+
+async def execute_integration(
+    provider: str,
+    setup: ExecuteIntegrationSetup | None,
+    arguments: ExecuteIntegrationArguments,
+) -> str:
+    match provider:
+        case "duckduckgo_search":
+            return await duckduckgo_search(arguments=arguments)
+        case "dalle_image_generator":
+            return await dalle_image_generator(setup=setup, arguments=arguments)
+        case "wikipedia":
+            return await wikipedia(arguments=arguments)
+        case "weather":
+            return await weather(setup=setup, arguments=arguments)
+        case "hacker_news":
+            return await hacker_news(arguments=arguments)
+        case _:
+            raise ValueError(f"Unknown integration: {provider}")

integrations-service/integrations/utils/integrations/__init__.py

@@ -0,0 +1,5 @@
+from .dalle_image_generator import dalle_image_generator
+from .duckduckgo_search import duckduckgo_search
+from .hacker_news import hacker_news
+from .weather import weather
+from .wikipedia import wikipedia