Terminology

This document explains the core terms and important concepts used in Magic.

Basic Concepts

Term	Description
Model	Models contain a large number of parameters and are trained on vast amounts of diverse data. You can use models to perform various tasks such as image generation, question answering, code generation, rewriting, etc.
Access Token	When calling Magic API and SDK, access tokens are required to verify user identity and authorize access. Users can set different permissions for each access token to achieve secure resource access.
Token	In large language models, tokens are the basic units of text processing. Models typically break down input text into a series of tokens, then process and analyze these tokens. Tokens can be words, characters, subword segments, or other forms of text fragments. The specific segmentation method depends on the tokenization algorithm used by the model, so token calculation and processing methods may vary according to the model's specific architecture and design. When conversing with an AI assistant, the generated tokens are the sum of input tokens and output tokens.
Prompt	Natural language instructions that tell the model what task to perform. For example, "Translate the text into English". It is recommended to specify the model's role in the persona and response logic, design the response language style, and limit the model's answer scope to make conversations more aligned with user expectations.
Workflow	A workflow is a tool used for planning and implementing complex functional logic. You can design complex multi-step tasks by dragging different task nodes, improving construction efficiency in complex business scenarios.
Knowledge	Knowledge base is the AI assistant's private knowledge collection, used to store professional domain knowledge data, solving the problem of insufficient professional domain knowledge in general large models, and improving the professionalism and accuracy of AI assistant responses.
Memory	Memory refers to the ability of artificial intelligence to store, retrieve, and utilize information over time, improving model performance by providing more accurate and contextually relevant responses. Magic provides a set of memory functions, including variables, databases, and long-term memory, to meet different use cases. Based on these memories, personalized responses can be provided to enhance user experience.

Node Names

Category	Node Name	Node Description
Basic	Start Node	When the following events are triggered, the process will start execution from this module
Basic	Message Reply	Reply or respond with specified content to the user
Basic	End Node	The final node of the process, used to return result information after workflow execution
Basic	Wait	The process will wait at this node for the user's next operation
Large Model	Large Model Call	Call large language models, use variables and prompts to generate responses
Large Model	Intent Recognition	Large model recognizes the actual intent of input content based on the input
Operations	Selector	Connects multiple downstream branches, runs the "corresponding" branch if set conditions are met, otherwise runs the "else" branch
Operations	Code Execution	Write code to process input variables, then generate return values through output
Operations	HTTP Request	Send requests to external HTTP services according to set parameters and obtain response data
Operations	Sub-process	Assign some functional modules to sub-processes for orchestration, avoiding the main process becoming too large
Operations	Tool	Reusable tool processes, transferring some process functionality to tools for implementation
Operations	Personnel Retrieval	Search for personnel information that meets conditions, only supports viewing personnel visible to the conversationalist and related information
Operations	Loop	Repeatedly execute a series of tasks by setting loop count and logic
Operations	Knowledge Retrieval	Perform knowledge retrieval on input keywords, return matching relevant content
Operations	Cloud Document Parsing	Output content of specified cloud documents in Markdown structure through nodes
Operations	Image Generation	Generate images through text descriptions
Operations	Create Group Chat	Create group chats containing specified group members
Data Processing	Historical Message Query	Query historical messages according to specified conditions
Data Processing	Historical Message Storage	Custom memory storage, freely choose to store any content
Vector Knowledge Base	Vector Storage	Store in knowledge base in fragment form
Vector Knowledge Base	Vector Search	Perform similarity matching from top to bottom in multiple knowledge bases, output fragments of specified similarity and quantity
Vector Knowledge Base	Vector Deletion	Perform similarity matching from top to bottom in multiple knowledge bases, output fragments of specified similarity and quantity
Vector Knowledge Base	Vector Knowledge Base Matching	Match related vector knowledge bases through query conditions
Persistent Database	Data Storage	Store persistent data
Persistent Database	Data Loading	Read persistent data
Variables	Variable Save	Update if variable exists, create new if it doesn't exist
Magic Tables	Add Record	Add records to specified data table
Magic Tables	Modify Record	Modify records in specified data table
Magic Tables	Query Record	Query row records in data table according to specified conditions
Magic Tables	Delete Record	Delete row records in data table according to specified conditions
File Related	Document Parsing	Extract file text content, output as text to the next node for use
File Related	Spreadsheet Parsing	Extract file text content, output as text to the next node for use
Text Related	Text Splitting	Split long text according to established strategies, strategy selection will be opened in the future