Skip to content

Terminology

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

Basic Concepts

TermDescription
ModelModels 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 TokenWhen 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.
TokenIn 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.
PromptNatural 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.
WorkflowA 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.
KnowledgeKnowledge 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.
MemoryMemory 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

CategoryNode NameNode Description
BasicStart NodeWhen the following events are triggered, the process will start execution from this module
BasicMessage ReplyReply or respond with specified content to the user
BasicEnd NodeThe final node of the process, used to return result information after workflow execution
BasicWaitThe process will wait at this node for the user's next operation
Large ModelLarge Model CallCall large language models, use variables and prompts to generate responses
Large ModelIntent RecognitionLarge model recognizes the actual intent of input content based on the input
OperationsSelectorConnects multiple downstream branches, runs the "corresponding" branch if set conditions are met, otherwise runs the "else" branch
OperationsCode ExecutionWrite code to process input variables, then generate return values through output
OperationsHTTP RequestSend requests to external HTTP services according to set parameters and obtain response data
OperationsSub-processAssign some functional modules to sub-processes for orchestration, avoiding the main process becoming too large
OperationsToolReusable tool processes, transferring some process functionality to tools for implementation
OperationsPersonnel RetrievalSearch for personnel information that meets conditions, only supports viewing personnel visible to the conversationalist and related information
OperationsLoopRepeatedly execute a series of tasks by setting loop count and logic
OperationsKnowledge RetrievalPerform knowledge retrieval on input keywords, return matching relevant content
OperationsCloud Document ParsingOutput content of specified cloud documents in Markdown structure through nodes
OperationsImage GenerationGenerate images through text descriptions
OperationsCreate Group ChatCreate group chats containing specified group members
Data ProcessingHistorical Message QueryQuery historical messages according to specified conditions
Data ProcessingHistorical Message StorageCustom memory storage, freely choose to store any content
Vector Knowledge BaseVector StorageStore in knowledge base in fragment form
Vector Knowledge BaseVector SearchPerform similarity matching from top to bottom in multiple knowledge bases, output fragments of specified similarity and quantity
Vector Knowledge BaseVector DeletionPerform similarity matching from top to bottom in multiple knowledge bases, output fragments of specified similarity and quantity
Vector Knowledge BaseVector Knowledge Base MatchingMatch related vector knowledge bases through query conditions
Persistent DatabaseData StorageStore persistent data
Persistent DatabaseData LoadingRead persistent data
VariablesVariable SaveUpdate if variable exists, create new if it doesn't exist
Magic TablesAdd RecordAdd records to specified data table
Magic TablesModify RecordModify records in specified data table
Magic TablesQuery RecordQuery row records in data table according to specified conditions
Magic TablesDelete RecordDelete row records in data table according to specified conditions
File RelatedDocument ParsingExtract file text content, output as text to the next node for use
File RelatedSpreadsheet ParsingExtract file text content, output as text to the next node for use
Text RelatedText SplittingSplit long text according to established strategies, strategy selection will be opened in the future

Released under the Apache 2.0 License