Atomic State

AtomicState

An atomic state is a state executing real tasks, which usually are small functional modules, such as LLM module and TTS module.

In MyShell bot, an entered atomic state usually means a sent message with buttons if specified.

AtomicState

Field's Name	Type (Required/Optional)	Description	Example
id	string (Optional)	Globally unique identifier.	"1234_5678_9101"
type	"state" (Optional)	Specifies that this is an atomic state.	"state"
properties	Object (Optional)	Static config. Similar to metadata.
properties.is_final	boolean (Optional)	Flag to check if the current state is final. Default to false.	true
properties.cache	boolean (Optional)	Flag to enable cache mode. When set to true, the state will run only once and store the result. Default to false.
inputs	Object (Optional)	Expected inputs, including user inputs and other states' outputs.
inputs.[input_name]	Input or Expression (Required)	Specification about each input.	{"type" : "IM", "user_input": true}
tasks	SupportModule[] (Optional)	Tasks executed in written order, involving no control ability such as if/else and loop. Each element is configuration for supported modules, including LLMModule, LLMFunctionModule and TtsModule for now.
outputs	Object (Optional)	Output values, including state's outputs and modification to parent automata's context, if it has a parent automata.
outputs.[output_name]	Variable or Expression (Required)	Usually its an expression evaluating an intermediate variable.	{"context.prompt" : "{{reply}}"}
render	RenderConfig (Optional)	Controls how to display the result to the user.	{"text": "{{reply}}", "buttons": ["context": "Chat", "on_click": "start_chat"]}
transitions	Object (Optional)	Indicates how state flows responding to user action. transitions's keys are events which the state will handle, including events triggered by itself or its children states if its children states don't handle this event. We treat both actual happened events and user intents as events. AtomicState can trigger special events, including CHAT, ALWAYS
transitions[event]	Transition (Required)	It could simply a target state name, or a TransitionCase object.	{"start_chat" : {"target": "chat_page"}}

Example:

{
  "inputs": {
    "user_input": {
      "type": "IM",
      "user_input": true
    }
  },
  "tasks": [
    {
      "module_type": "LLMFunctionModule",
      "module_config": {
        "model": "gpt-35-turbo-16k",
        "system_prompt": "{{context.system_prompt}}",
        "user_prompt": "{{context.prefix_prompt}}\n{{user_input}}\n{{context.suffix_prompt}}",
        "function_name": "reply_to_user",
        "function_description": "Generate reply to user.Always use this function.",
        "function_parameters": [
          {
            "name": "reply",
            "type": "string",
            "description": "This is your reply to user. AWAYS IN ENGLISH"
          }
        ]
      }
    },
    {
      "module_type": "TtsModule",
      "module_config": {
        "content": "{{reply}}",
        "tts_id": "2",
        "output_name": "reply_voice"
      }
    }
  ],
  "render": {
    "text": "{{reply}}",
    "audio": "{{reply_voice}}",
    "buttons": [
      {
        "content": "Return",
        "description": "",
        "on_click": "return"
      }
    ]
  },
  "transitions": {
    "CHAT": {},
    "create_scenario": "new_scenario"
  }
}

Inputs

Each Input signifies that a variable is required for the state to function, and it is typically provided by user input. If there is user input whose type is not "IM", then a transition to this state would prompt the opening of a modal, allowing the user to complete the input form.

The primary distinction between Input and Output in this context is that Input generally includes additional fields that govern how the user should complete the input form. These fields could include things like data validation rules, default values, or specific instructions for the user, which provide guidance on the expected form and content of the input. Output, in contrast, typically refers to the information that is conveyed back to the user after processing their input or completing a particular state transition.

Input

Field's Name	Type (Required/Optional)	Description	Example
type	"text" | "image" | "audio" | ”IM" (Required)	The type of a given variable. "IM" refers to input that originates from the user's bot Instant Messaging interface.	"text"
value	string (Optional)	The value of a given variable. The type of the value depends on the type field. For text, value could be any string. For image, audio, value is string of URL. ForIM, value is usually undefined. If the transition to the state specifies target_inputs, value won’t be overridden.	"Hello, World!" or "https://example.com/audio.mp3"
default_value	string (Optional)	The default value of an input variable, which is used when no value is given. The type is same as value. If the transition to the state specifies target_inputs, default_value will be overridden.	"{{prompt}}"
user_input	Boolean (Optional)	An optional flag that indicates whether the input should come from the user. Defaults to false.	true
name	string (Optional)	An optional string that serves as a label for the form input.	"username"
description	string (Optional)	An optional string that serves as a description for the form input.	"Input your username here."
choices	string[] (Optional)	Allow users to select the value from given choices.	["GPT", "Gemini"]
validations	Validation[] (Optional)	Allow users to specify how this input should be validated. The validators are processed one by one. The first failed validation will produce corresponding error message. If not specified, it is equivalent to set to [{"required": true}]	[{"required": false}, {"max_length": 100}]

Validation

Field's Name	Type (Required/Optional)	Description	Example
required	boolean (Optional)	Defaults to true. If set to false, the input is optional.
max_length	number (Optional)	The character limitation for a text input. Defaults to 1500.	"Hello, World!" or "https://example.com/audio.mp3"
max_file_size	number (Optional)	The file size (in bytes) limitation for file-type inputs including image, audio, video and file.	10 * 1024 * 1024
max_number	number (Optional)	The max number for number and integer inputs.	1500
min_number	number (Optional)	The min number for number and integer inputs.	-1
error_message	string (Optional)	(Coming soon) The error message the user would see when validation fails.

Tasks

We support LLM module, LLM Function module and TTS module for now. More kinds of modules and customized modules are coming very soon.

To fully understand the configuration of these modules, you should refer to the Modules section where detailed information about each module, including their inputs, outputs, and functionality, is provided.

Important Update: In previous versions, we supported the Object type for defining tasks. Please be aware that the execution order cannot be guaranteed for the Object type and it will become deprecated in a future release. It is recommended to transition to using the Array type to ensure the execution order of tasks.

Outputs

Currently, the use of context is necessary to store any output variables, as they are fundamentally kept within the parent automata's scope. However, in a few days, we will introduce support for isolated output variables that can be accessed through a prefix tied to the state name.

Variable is a subset of Input, only including type and value fields.

Variable

Field's Name	Type (Required/Optional)	Description	Example
type	"text"	"image"	"audio"
value	string (Optional)	The value of a given variable. The type of the value depends on the type field. For text, value could be any string. For image, audio, value is string of URL. ForIM, value is usually undefined. For expression, value is Expression.	"Hello, World!" or "https://example.com/audio.mp3"

Render

RenderConfig is responsible for defining how a bot presents the results executed by a state to the user. It can specify content of the result, whether it's a text message, an audio message, or interactive buttons that facilitate transitions to different states or prompt further user interaction.

RenderConfig

Field's Name	Type (Required/Optional)	Description	Example
text	string (Optional)	The text that would be displayed on the render.	"This is the render text."
image	string (Optional)	A URL string to an image that would be displayed on the render.	"https://example.com/image.jpg"
audio	string (Optional)	A URL string to an audio file that would be played on the render.	"https://example.com/audio.mp3"
buttons	Button[] | Button[][] (Optional)	An array or a matrix of Button objects that may appear on the render for interactive purposes. If set as a matrix, each array of buttons will be displayed in a single row.

Button

Field's Name	Type (Required/Optional)	Description	Example
content	string (Required)	The visible text displayed on the button.	"Click Me"
description	String (Optional)	A tooltip that appears when you hover over the button.	"This button triggers the next part of the process"
on_click	string | Event (Required)	Defines the event that should be triggered when the button is clicked. You can pass data to Transition if needed.	"start_chat"

Event

Field's Name	Type (Required/Optional)	Description	Example
event	string (Required)	The event name.	"create_page"
payload	Object (Optional)	When triggering an event, you may pass data to the TransitionCase by setting values in payload. In TransitionCase , passed data can be accessed by the prefix payload. .