surfaice-assistant 0.3.4

No project description provided

Assistant Processor Guides

Warning

The file processor_2.py is a temporary file used in the current branch and will be renamed to processor.py. Please be aware that any references to processor_2.py should be updated to processor.py once the renaming is complete.

Installation

Poetry Installation from Git

To install this package directly from Git using Poetry, add the following to your pyproject.toml:

[tool.poetry.dependencies]
surfaice-assistant = { git = "https://github.com/gdml/surfaice-assistant.git", branch = "function_calling" }

or locally:

[tool.poetry.dependencies]
surfaice-assistant = { path = "path/to/surfaice-assistant"}

Usage

Basic Setup

from surfaice_assistance.processor2 import AssistantProcessor   

class YourClass:
    ...

    @property
    def assistant_processor(self) -> AssistantProcessor:
        return AssistantProcessor(api_key=self.api_key, provider=self.provider.value)
    ...

    def create_assistant(self) -> AssistantCreateResponse:
        return self.assistant_processor.create_assistant(
            name="Your Assistant Name",
            description="Your Assistant Description",
            instructions="Your Assistant Instructions",
            model="gpt-4o",
            tools=list[dict],
        )   

    def get_assistant(self) -> AssistantGetResponse:
        return self.assistant_processor.get_assistant(
            assistant_id="your_assistant_id",
        )
    
    def update_assistant(self) -> AssistantUpdateResponse:
        return self.assistant_processor.update_assistant(
            assistant_id="your_assistant_id",
            name="Your Assistant Name",
            description="Your Assistant Description",
            instructions="Your Assistant Instructions",
            model="gpt-4o",
            tools=list[dict],
        )
    
    # ... other methods

Adding a New Implementation

When adding a new assistant implementation to the system, follow these steps:

1. Create a new implementation class that inherits from AssistantInterface:

   • Create a new file in the implementations/ directory
   • Implement all abstract methods from AssistantInterface
   • Follow the response models pattern using Pydantic models
   • Add import to __all__ in implementations/__init__.py

2. Register the implementation in processor_2.py:

   from implementations.your_provider import YourProviderImpl
   
   processor_implementations: dict[str, Union[Type[OpenAIAssistantImpl], Type[YourProviderImpl], Any]] = {
       OPENAI: OpenAIAssistantImpl,
       "your_provider": YourProviderImpl,  # Add your implementation here
   }
   

3. Define a constant for your provider name at the top of processor_2.py:

   YOUR_PROVIDER = "your_provider"
   

4. Ensure your implementation:

   • Uses the standard response models from abstract_provider.assistant
   • Handles retries and error cases consistently
   • Includes proper logging
   • Has appropriate type hints
   • Follows the existing pattern for async methods

Implementation Requirements

Your implementation class must:

1. Inherit from AssistantInterface
2. Implement all abstract methods defined in the interface
3. Use the standard response models:
   • AssistantCreateResponse
   • AssistantGetResponse
   • ThreadGetResponse
   • MessageCreateResponse
   • etc.

Example Implementation Structure

from abstract_provider import AssistantInterface
from abstract_provider.assistant import (
    AssistantCreateResponse,
    AssistantGetResponse,
# ... other response models
)
class YourProviderImpl(AssistantInterface):
    def init(self, api_key: str) -> None:
    # Initialize your provider's client
        pass
    async def create_assistant(
        self,
        name: str,
        description: str,
        instructions: str,
        model: str,
        tools: list[dict],
        max_retries: int = 5,
        ) -> AssistantCreateResponse:
        pass

Reference Implementation

See the OpenAI implementation (implementations/openai_provider.py) for a complete example of how to structure your implementation.

Guidelines for Response Models

• Use Pydantic models for response models
• Follow the response models pattern using Pydantic models
• Add the response models to the implementations.your_provider module
• Add the type of your models to unions in abstract_provider.assistant models
• Example:

  from pydantic import BaseModel
  class YourResponseModel(BaseModel):
      # Define your model fields here
      success: bool   
      your_field: Optional[YourType] = None
      error: Optional[str] = None