Skip to main content
The AI Agent Widget is your context-aware assistant, accessible via the Widget Bar. Unlike general-purpose AI, the Custral Agent is deeply integrated with your organization’s data, allowing it to provide precise answers and execute complex workflows.

​
Interaction Modes

​
Widget Bar Chat

The primary way to interact with the Agent. It lives in the global dock, allowing you to ask questions or issue commands without leaving your current page.

​
Inline AI Actions

Many record fields and surfaces have “AI Spark” icons. Clicking these allows the Agent to perform localized tasks, such as summarizing a specific note or reformatting a text field.

​
Agentic Workflows

The Agent can be triggered as part of a Workflow to perform background data synthesis or automated customer outreach.

​
Key Capabilities

  • Natural Language Querying: Ask questions about your records, such as “Who are our top 5 customers by revenue?” or “Summarize the last three calls with Acme Corp.”
  • Task Automation: Command the agent to perform actions, like “Create a follow-up task for Tuesday” or “Draft a summary email for this deal.”
  • Contextual Awareness: The agent knows which page or record you are currently viewing and uses that information to provide more relevant answers.
  • Data Synthesis: Extract insights from multiple sources, including conversations, meeting notes, and structured record fields.

​
How to Use the Agent

The AI Agent is typically accessible via a chat icon in the Custral sidebar or footer. You can interact with it in several ways:
  1. Direct Chat: Type your questions or commands into the chat box.
  2. Contextual Actions: Use pre-defined “Quick Actions” that appear based on the record you are viewing.
  3. Voice Commands: (If enabled) Use voice-to-text to interact with the agent hands-free.

​
Configuration

​
Privacy & Security

The AI Agent respects all existing Access Control Lists (ACLs). It will never surface data to a user who isn’t authorized to see it, and your organization’s data is never used to train models for other customers.

​
Example Prompts

  • “What are the next steps for the ‘Project Phoenix’ onboarding?”
  • “Find all support tickets related to ‘API Timeout’ from the last 24 hours.”
  • “Summarize the sentiment of the last 5 conversations with our Enterprise accounts.”
  • “Help me draft a renewal proposal for the ‘Skyline Systems’ account based on their usage data.”