Skip to main content

What is a Flow?

Concept​

Flow is set of sequencially linked nodes that execute to create a response for a roleplay session. A flow has to be connected from the start node to the end node (with the agent nodes in between) for it to execute properly.

Concept

Basic Components​

Node​

A node is the basic execution unit of a flow. Nodes come in three types:

  • Start node: The start node is where the flow begins. .
  • Agent node:
    • An agent calls an AI model and obtains an output that can be used later in other agents or in response design.
    • Agents can take on different responsibities depending on the nature of the roleplay and the AI model used.
  • End node:
    • The end node is where the flow ends.
    • In the end node, users can use the response design feature to customize how the AI reponse in displayed during the roleplay session.

Node

Edge​

An edge is a connection line between nodes. Nodes are executed sequentially according to the edges connected from the start node to the end node. Edges serve the following roles:

  • Defines Execution Order: Determines when each node will be executed.
  • Controls Data Flow: Determines the path through which context/information is processed.

Variables​

Each node can utilize variables stored in the context using template syntax. Variables can be referenced in the form {{variable_name}} and used in node input fields or prompts.

For example:

  • {{char.name}}: The name of the character currently being referenced or taking action in a roleplaying sequence.
  • {{user.description}}: The description of the character controlled by the user in a roleplay.
  • {{session.entries}}: A list of all retrieved character and plot entries.

There are two types of variables in astrsk.ai:

  • (System) variables: Basic variables that is provided by the application. These can be used in any flow or card.
  • Agent output: Output data from an agent node. Agent outputs are restricted to use within the flow where they were generated.

Variables

Example​

Here's a simple flow example using two agents:

Start Node → Agent 1 (Input Analysis) → Agent 2 (Response Generation) → End Node

Execution Process:

  1. Start Node: Initial context setup (character information, conversation history, etc.)
  2. Agent 1 (Input Analysis): Analyzes user input to generate {{user_intent}} and {{context_summary}} variables
  3. Agent 2 (Response Generation): Uses Agent 1's analysis results along with {{char.description}}, {{char.example_dialog}}, etc. to generate {{char_response}} variable
  4. End Node: Outputs {{char_response}} content as character message

Context accumulates at each step, and the next node can utilize the results from previous nodes.

Example

References​

For a complete list of available variables and detailed information about each variable, please refer to the Variable Reference. Here you can find data types, usage methods, and examples for each variable.