On devices that support Google Assistant,
Natively built in an officially supported third-party platform known as Actions On Google (AoG).

As a user requests to interact with a action on this third-party platform, the system will automatically search for the corresponding action to be serived. Once the platfoem found, the control of the response will be handed over to the third-party action. In this point, the Google Assistant's role has shifted to an intermediary that performs speech recognition and presents third-party action content.

In practice, the entire interaction process is information transfer through API in JSON format.

As the user interact with the action, the following pipeline with be practice
Until the request of the user is fulfilled or the action is interrupted, each round of the conversation is perform in this way:

• 1. User asked to the Assistant
• 2. Assistant transfer the (recognized) sentences send to the Action
• 3. In the structure of the action, we use DialogFlow which containing the pretrained modle to judged the intent of the user
• 4. To perform the complex responds, the DialogFlow narrow down the sentence and send the coresponding parameters to the Fulfillment where will perform data fetching or API call
• 5. Fulfillment return the relative Intent responds to the Assistant
• 6. Assistant recived the responds and show it to the user