Creating an Assistant
Walk through the process of creating an Assistant via Magic Studio
Last updated
Walk through the process of creating an Assistant via Magic Studio
Last updated
Signup to Conva.AI via Magic Studio
Go to https://studio.conva.ai
Use the "Login with Google" option for a faster signup process.
Magic Studio creates a unique OrganizationID for every user who signs up, The Organization Name is the Display Name of that ID. You can accept the default suggestion or give a proper name.
If you want multiple users to belong to the same Organization so that they can share the same Assistant, please send an email to 42@slanglabs.in
Click on the "Create Assistant" button. This will trigger the Assistant creation wizard.
In this step, you should give some details of the app to where the Assistant will be integrated into. This will be used by Conva.AI to ensure that it understands the larger scope and categories supported by the Assistant. Magic Studio provides an easy way to get this done. Just give the best URL that reflects your app - usually your Playstore or Appstore URL or your home page URL. Magic Studio will then automatically generate the required details. You can edit the information if needed.
Sometimes this operation might not generate all the details. In such cases, just click "Re-generate"
While some of the fields in App Details section might be obvious (like App Name and App Description), others might need some explanation
| Name of the field | Description of the field |
|---|---|
Project Categories | This is used to ground the things that the Assistant is supposed to know about and is usually the list of product categories supported by the app and/or its features. |
Regions | Which regions does the app operate in. If there is no region restriction, just mention "Global" |
At any point in the flow, you can go back all the way to the end and cancel the Assistant Creation. Magic Studio will automatically save a draft copy and you can resume the creation process anytime in the future.
The next step is to add one or more Capabilities for the Assistant. Remember Capability represents a specific task that the Assistant needs to handle in order to trigger something inside your app or answer a question. These are the foundational units that make up your Assistant.
Refer to the Capabilities section in the overview page for more details about it
By default, Magic Studio will add two default Capabilities that handles some of the most common scenarios.
Casual Conversations - This is to handle cases where users are doing chit-chat with the Assistant.
Unsupported Conversations - This is to prevent the Assistant from answering things that is outside the scope of the App and outside the list of enabled Capabilities
Creators cannot remove "Unsupported Conversation". But you can disable it later in the edit screen
Click on the "Create New Capability" to add your own Capability. This will trigger the "Capability Creation" flow
"Domains": This is the selected domain of the app. Its expected to be one of the 32 app types that Google Play Store uses to classifies all its apps.
Users can either create their own Custom Capability or chose from the list of standard Capabilities that Magic Studio supports that can be imported into their Assistants.
To create your own Capability, you need to write a brief description of what the Capability should do.
Magic Studio will automatically generate the Capability blueprint from this description
Refer to the "Capability Configuration" section in the Conva.AI Overview page for more details of the contents from this Capability blueprint
Feel free to edit what Magic Studio has generated to better suit your needs.
Once you are happy, you can go ahead and click "Save".
You can then repeat this process of adding more Capabilities.
The next step is to build the Assistant. This is the step where Magic Studio will generate the final set of prompts that will eventually be used at runtime (well almost final prompts).
Magic Studio comes with a Capability Compiler that takes the Capability blueprints and generates the prompts.
This process typically takes a few minutes as it internally uses a powerful but relatively slower LLM
The next steps it to test out the Assistant.
Enter the Playground!
