Connector Configuration

This article aims to explain the configuration structure of Custom Connector, and you can find Python examples of Chat, Text Completion and Embedding in Custom connector examples.

Data Structures of Messages and Responses

For how to use these data structures, please refer to Custom Connector Overview

from pydantic import BaseModel

class Message(BaseModel):
    role: str
    content: str

class RequestSettings(BaseModel):
    temperature: float
    maxResponseTokens: int
    topP: float
    frequencyPenalty: float
    presencePenalty: float

class CopilotChatRequest(BaseModel):
    message: str
    history: List[Message]
    settings: RequestSettings

class CopilotTextRequest(BaseModel):
    message: str
    settings: RequestSettings

class CopilotMessageResponse(BaseModel):
    content: str
    isError: bool

Connector Properties

Here are some definitions of connectors that determine how they are displayed, categorized, and launched.

Property name	Required?	Description	Example value
id	Y	Identifier of the connector, it should be unique	com.richasy.connector
name	Y	Name of the connector, it will be displayed on the UI	My Connector
base_url	Y	API service address, must be http/https protocol	http://localhost:8000
config_path	N	If the connector requires manual configuration by the user, such as adding keys, etc., the relative path of the configuration file can be written here	config.json
readme	N	If there is a description file, please fill in this field. After the user imports, the application will automatically open the description file	README.txt
execute_name	N	If the API service entry is an executable file, please fill in this field	api.exe
script_file	N	If the API service needs to be started by running a script, please fill in this field. The application will run this script through powershell	run.ps1
script_command	N	If the API service needs to be started by a command, please fill in this field. The application will execute this command through powershell	python api.py
features	Y	List of Features, please refer to Feature	[{...}, {...}]

NOTE
execute_name, script_file and script_command are the main ways to start the API service, and a connector can only choose one of them.
If your API service is deployed in the cloud and the data structure meets the expectations of Fantasy Copilot, you can omit these fields and fill in your deployment URL in base_url.

Feature

Feature defines the features supported by the connector.

Property name	Required?	Description	Example value
type	Y	Feature type, currently supports chat, text-completion and embedding	chat
endpoints	Y	Defined API endpoints list, see Endpoint for specific data structure	[{...}, {...}]

About Feature Type

• chat: The connector supports conversation. It can be used in session scenarios.
• text-completion: The connector supports text completion. It is also applicable to the use scenarios of chat, but if you need to use knowledge base or semantic skills, you must add text completion model.
• embedding: The connector supports text vectorization (embedding). If you need to use knowledge base, you must add a connector that supports embedding.

Endpoint

Endpoint refers to our API endpoints.

Property name	Required?	Description	Example value
type	Y	Defines the endpoint type, see Endpoint type for supported endpoint types	chat-rest
path	Y	Endpoint path relative to base_url	/chat

Finally, the application will splice base_url and path according to the current settings to form a complete url such as http://localhost:8000/chat.

Endpoint type

Type	Belonging feature	Description
chat-rest	chat	It represents a chat POST request, and the service will return a complete text response
chat-stream	chat	The service will generate and return one character at a time in the form of SSE (Server-Send Events)
chat-stream-cancel	chat	Inform the service to stop executing chat-stream
text-completion-rest	text-completion	The service will return a complete response after text completion
text-completion-stream	text-completion	The service will generate and return one character at a time in the form of SSE (Server-Send Events)
text-completion-cancel	text-completion	Inform the service to stop executing text-completion-stream
embedding-rest	embedding	Input text and generate vectors, then return the result

Full Configuration

{
    "id": "com.richasy.connector",
    "name": "My Connector",
    "base_url": "http://127.0.0.1:8000",
    "config_path": "config.json",
    "readme": "README.txt",
    "script_file": "If needed",
    "script_command": "If needed",
    "execute_name": "If needed",  
    "features":[
        {
            "type": "chat",
            "endpoints":[
                {
                    "type": "chat-rest",
                    "path": "/chat"
                },
                {
                    "type": "chat-stream",
                    "path": "/chat-stream"
                },
                {
                    "type": "chat-stream-cancel",
                    "path": "/chat-stop"
                }
            ]
        },
        {
            "type": "text-completion",
            "endpoints":[
                {
                    "type": "text-completion-rest",
                    "path": "/text"
                },
                {
                    "type": "text-completion-stream",
                    "path": "/text-stream"
                },
                {
                    "type": "text-completion-stream-cancel",
                    "path": "/text-stop"
                }
            ]
        },
        {
            "type": "embedding",
            "endpoints": [
                {
                    "type": "embedding-rest",
                    "path": "/embedding"
                }
            ]
        }
    ]
}