​
Working with Assistants

Assistants are the core of Kosmoy Studio’s no-code AI application development. They are AI-powered entities designed to perform specific tasks or fulfill particular business objectives. This section guides you through creating, configuring, deploying, testing, and monitoring your Assistants.

​
Assistant Concepts

  • No-Code AI: Assistants are built using a visual, intuitive interface, requiring no coding knowledge.
  • Templates: Kosmoy Studio provides pre-built templates to help you quickly create common types of Assistants. Current templates include:
    • Receptionist Assistant: Handles reception-related tasks (e.g., scheduling meetings). Requires a Google or Outlook Calendar integration.
    • HR Engagement Analyst: Provides insights and analysis related to human resources.
    • Wealth Management Assistant: Assists with financial planning and investment tasks.
    • Personal Assistant: A general-purpose assistant for various personal tasks.
    • Sales Assistant: Supports sales-related activities.
    • Customer Support Assistant: Helps with customer service inquiries.
  • Private vs. Public:
    • Private Assistants: Accessible only to users within your Kosmoy organization via Kosmoy Chat.
    • Public Assistants: Designed to be accessible to the general public through configured Communication Channels (e.g., WhatsApp via Twilio).
  • Data Channels: Assistants can be connected to data channels.
  • Gateways: Assistants can be connected to Gateways

​
Creating an Assistant

  1. Navigate to Assistants: From the Kosmoy Studio home page, click on the “Apps” menu in the left-hand navigation bar, then select “Assistants”.

  2. Add a New Assistant: Click the ”+ ADD” button located in the upper right corner of the Assistants section.

  3. Step 1: Set Assistant Properties:

    • Avatar: Select an avatar for your Assistant.
    • Name: Give your Assistant a unique and descriptive name.
    • Description: Provide a brief description of your Assistant’s purpose.
    • Template: Choose an Assistant template from the dropdown menu. Note: If you select the “Receptionist Assistant” template, you will be prompted to select a pre-configured Google or Outlook Calendar integration.
    • Scope: Select either “Private” or “Public” from the dropdown menu. This determines how the Assistant will be deployed and accessed.
    • Language: Select the primary language for your Assistant. The default is “auto,” which attempts to automatically detect the language. Other options include English, Spanish, Italian, and French.

  4. Click “Next”.

  5. Step 2: Set Assistant Data Access and Memory:

    • Data Channel (Optional): Select a pre-configured Data Channel from the dropdown menu to connect your Assistant to a data source.
    • Memory: Configure the Assistant’s memory settings:
      • Time Window: Define the length of time (e.g., minutes, hours, days) for which messages should be kept in memory. The default is 1 day.
      • Past Interactions: Specify the maximum number of past interactions (question-answer pairs) to retain within the time window. The maximum is 25, and the default is 10.
    • Conversation Logger (Optional): Enable this feature to record assistant interactions, which is helpful to perform a QA over the AI assistant performance.
      • Select SQL database with write rights enabled
      • Conversation Breakdown Strategy: Select the amount of time needed to pass from the last message received to consider the current conversation session finalized.
      • Decide if you would like to track user data or not

  6. Click “Next”.

  7. Step 3: Set Assistant Security Parameters: This step will be deprecated in the future.

    • Guardrails Panel (Optional): Enable the global toggle and select individual guardrails (Prompt Injection, Harmful Content, Toxic Language) to apply additional safety measures. Note: Guardrail configuration will eventually be handled exclusively at the Gateway level.

  8. Click “Next”.

  9. Step 4: Set Assistant Models and Prompt:

    • Model Panel:
      • Conversational Model: Select the primary model that will power the Assistant’s conversational abilities.
      • Instructional Model: Select the model used for instructional tasks or specific commands. NOTE: In the future the user will be prompted to select a gateway and, based on that selection, models and routers will be listed here to be selected by the user.
    • Assistant Prompt Panel: Define the core instructions and personality of your Assistant:
      • Role: Describe the Assistant’s role (e.g., “Helpful Customer Support Agent”).
      • Task: Specify the Assistant’s primary task (e.g., “Answer customer questions about products and services”).
      • Operating Principles: Define guidelines for the Assistant’s behavior (e.g., “Be polite and professional,” “Provide accurate information”).
      • Attitude: Describe the desired attitude of your assistant (e.g. “Friendly and easy-going”).
      • Welcoming Message: Set the initial message the Assistant will use to greet users.
      • Goodbye Message: Set the message the Assistant will use when ending a conversation.

  10. Click “Review”: Review all the Assistant’s configuration details.

  11. Click “Create”: Create the Assistant.

​
Assistant Cards

The Assistants section displays each created Assistant as a card. Each card shows:

  • Assistant Avatar: The selected avatar for the Assistant will be displayed as the card background.
  • Assistant Name: The name you assigned to the Assistant.
  • Description: The description you provided for the Assistant.
  • Scope Icon:
    • Globe Icon: Indicates a Public Assistant.
    • Person Icon: Indicates a Private Assistant.
  • Deployment Status:
    • Rocket Icon (no background): Indicates the Assistant is not deployed. Hovering over this icon will reveal a white background, indicating it’s clickable.
    • Rocket Icon (with background): Indicates that the assistant is deployed (Future Enhancement).
  • Test/Insights Icon:
    • Multi-panel Icon: If undeployed, clicking this icon initiates the Assistant Test creation workflow within the Quality Assurance section.
    • Check Mark Icon: If deployed, clicking this icon navigates you to the Assistant’s dashboard within the Insights section.
  • Edit Icon (Pencil): Click this icon to update the Assistant’s configuration (only if the Assistant is not deployed).
  • Delete Icon (Trash Bin): Click this icon to remove the Assistant (only if the Assistant is not deployed).

​
Managing Assistants

​
Updating an Assistant

You can update all configuration parameters of an Assistant only if it is not deployed. If the assistant is deployed, you will not be able to edit its parameters.

  1. Navigate to Assistants: Go to “Apps” > “Assistants”.
  2. Locate the Assistant Card: Find the card for the Assistant you want to update.
  3. Click the Edit (Pencil) Icon: This will open the update dialog.
  4. Modify Assistant Parameters: Update the Assistant’s configuration as needed.
  5. Click “Save”: Save the changes.

If you attempt to edit a deployed Assistant, a warning banner will be displayed, preventing modifications.

​
Removing an Assistant

You can remove an Assistant if it’s no longer needed. However, you cannot delete a deployed Assistant.

  1. Navigate to Assistants: Go to “Apps” > “Assistants”.
  2. Locate the Assistant Card: Find the card for the Assistant you want to remove.
  3. Click the Delete (Trash Bin) Icon: This will trigger a confirmation prompt.

If you attempt to delete a deployed Assistant, a modal will appear, preventing the deletion and explaining that the Assistant is in use.

  1. Confirm Deletion: Confirm that you want to delete the Assistant.

Warning: Deleting an Assistant is a permanent action and cannot be undone.

​
Deploying Assistants

The deployment process differs depending on whether you are deploying a Private or Public Assistant.

​
Deploying Private Assistants

  1. Locate the Assistant Card: Find the card for the undeployed Private Assistant you want to deploy.
  2. Click the Rocket Icon: This initiates the deployment process.
  3. Select Target Users: Select the specific Kosmoy users (within your tenant) who should have access to this Assistant in Kosmoy Chat.
  4. “Review” and “Deploy”

Once deployed, the selected users will see the Assistant listed in their Kosmoy Chat interface, allowing them to interact with it directly.

​
Deploying Public Assistants

  1. Locate the Assistant Card: Find the card for the undeployed Public Assistant you want to deploy.
  2. Click the Rocket Icon: This initiates the deployment process.
  3. Select Communication Channel: Choose a pre-configured Communication Channel from the dropdown menu. Currently, WhatsApp (via Twilio) is supported.
  4. “Review” and “Deploy”

Prerequisites:

​
Testing Assistants

Before deploying your Assistants, it’s highly recommended to thoroughly test them using the Assistant Tests feature in the Quality Assurance section. This allows you to interact with your Assistant in a controlled environment and refine its configuration. See Assistant Tests for more details.

​
Monitoring Assistants

After deploying your assistant, you would probably like to track its performance. See Dashboard for more details.

Introduction to Apps in Kosmoy StudioCoded Apps