Skip to main content

Introduction

An application (or app) wraps together the configuration for different components available in the SERMAS Toolkit.

Application details​

An application is composed of base information such as

  • appId A unique identifier
  • name The name for the app
  • description Description text for the app
  • public flag if the application can be listed (without authorization)

Modules​

Defines the modules the application can access within the Toolkit.

Each module definition will enable client credentials to be used for JWT access token retrieval.

The client will use the moduleId as client id and secret as client secret. If secret is not defined, it will be generated randomly.

The client can then be used to access the Toolkit API to interact with the modules specified in the supports list.

An example of a module definition follow:

- moduleId: <id of the module>
  secret: <module secret>
  supports:
  - dialogue
  - detection
  - ui
  - session
  - platform

Platform-level modules​

While some modules are natively part of the Toolkit, developers may add new modules by providing the details in the config field of the definition.

This will lookup and proxy requests toward an OpenAPI / AsyncAPI service and expose the signature as part of the main Toolkit API.

Repository​

The repository configuration collects configuration defining specific assetts to be used by the application

Avatars​

Define an avatar, which is a virtual representation of an Agent.

The relevant fields follows:

  • id Unique id of the avatar, used to select it in the settings
  • name Friendly name of the avatar
  • path 3D object asset path (suppoerted format are glb or fbx), loaded via the assets API e.g. avatars/avatar1.glb
  • gender: M or F, used in answer generation and for TTS model selection.
  • prompt Dedicated prompt used to define, characterize or constrain the avatar personality
  • camera, cameraMobile defines the camera position in the kiosk to position the avatar.

Backgrounds​

Provides a background image in the kiosk.

  • id background identifier
  • path relative path to the assets API
  • name Friendly name of the backgroun

Documents​

List of documents to import in the RAG pipeline. The Toolkit supports only text files.

The text files are vectorized following three strategies that can be specified in the metadata.options.parser field.

  • sentence (default) chunck the file text by sentences and handle them as documents.
  • single-line assume each single line (\n) in a text file is a document
  • double-line assume each double line (\n\n) in a text file is a document

Clients​

Clients allow third-party access to the API. Each client defines a clientId and secret used to obtain a temporary JWT token.

The permissions defines the list of <resource>.<scope> that client is allowed to access.

As example, dialogue or dialogue.* allow to access all the dialoge related functionalities.

The list of resources and scopes can be found in the API at http://localhost:8080/api/swagger#/PLATFORM/getSettings assuming you have a local instance running.

Otherwise you can derive the list from the topics in the source code (the pattern is app/:appId/<resource>/<scope>)