Architecture¶

Goal¶

The user of the unique toolkit interacts with the unique platform via services, concretely the

Knowledge Base Service - For all operations regarding the knowledge base only
Chat Service - For all interaction with the chat interface
Agentic Table Service (Soon) - To steer the agentic table interface

and well-known clients such as the OpenAI Client.

These interfaces are to be changed only by extension and the clients guarantee that the latest generative ai features are available quicker.

Platform Domains¶

The following domains exist

Domain	Description	Used by
`app`	Useful when dealing with settings and when dealing with events coming from the platform	Stateless API and Dev Setup
`chat`	Useful when interacting with the chat interface	`ChatService`
`content`	Useful when interacting with content saved in the knowledge base and chat	`ChatService` and `KnowledgeBaseService`
`framework_utilities`	Utilities to use e.g. OpenAIClient or `Langchain` and potentially other frameworks
`agentic`	All definitions to rebuild orchestrators and tools fully compatible with Spaces 2.0	`tool_packages` and `orchestrator`

Resulting Service¶

The services bundle the platform domains into services that interact with a single entity or user interface.

Service	Description	Maturity
`ChatService`	All capability to interact with the chat UI	️ ⚙️
`KnowledgeBaseService`	All capabilities to interact with the knowledge base	🧰 ⚙️ ️
`AgenticTableService`	All capabilities to interact with the agentic table	🧪

Maturity Levels:

Experimental 🧪: actively developed
In progress 🧰: Used in production but being extended
Stable ️⚙️: Widely used in production

Some domains resulted in Services that will be slowly deprecated

Services	Description	Replacement	Status	Usage
`EmbeddingService`	Access to embedding models	Use `OpenAIClient` directly	To be moved soon	Small
`LanguageModelService`	Access to language models	Use `OpenAIClient` directly	Deprecated	Medium
`ShortTermMemoryService`	Access to memory	Use in `ChatService` directly	To be moved soon	Small

Services and Clients utilities are always directly importable from the unique_toolkit.

from unique_toolkit import ChatService, KnowledgeBaseService 
from unique_toolkit import get_openai_client, get_async_openai_client, get_langchain_client

Data Objects¶

The DTOS and other data objects are implemented via pydantic which handles validation as well as name transformations between the platform and the python toolkit. The data objects can be found in the schema.py files and changed only by extension.

Underlying functions¶

Functions using the unique_sdk are defined in the functions.py files and might be changed as the platform evolves following usual deprecation patterns. Defining functionality per platform domains helps to test the functionality in a simpler fashion.