Skip to main content
Version: 6.4.0

Introduction to Agents

This page describes what SENSA Fabric Agents are and how they work. First read the Introduction to Fabric if you are familiarizing yourself with the Fabric toolset for the first time.

What are Agents?​

Agents are deployable applications that aggregate the logic, data, and models required to implement an AI use case. Agents are composed of one or more Skills (distinct units of functionality that provide actions), inputs, and outputs. Inputs and outputs take the form of service messages that provide a data structure and instructions to the Skill. Skills and service messages are wired together to form Agents.

Example Agent Use Case​

The example below is a simplification for the purpose of introduction.

Use Case Goals:

Customer required an Agent that could compare insurance claim forms to flag suspicious similarities and expose potential fraud. This AI-driven process saved many human-hours of work, improve accuracy, and save millions of dollars in avoiding payment for fraudulent claims.

Agent Skills:

  • Request documents from data source
  • "Read" document images
  • Compare images for similarities
  • Flag suspicious records for human review
  • Return results

Agent Lifecycle​

This section describes the general Agent lifecycle. For details and links to process documents for each part of the Agent lifecycle.

  1. Define the business use case for developing the Agent and architect it.
  2. Develop and deploy ML models in any IDE.
  3. Curate the data for training, testing, and deploying the models and save it a database or S3 bucket accessible to your Cortex cluster.
  4. Configure secrets required for access to the data content curated and the connections to Data Sources if Profile will be used in your agents.
  5. (Optional) Configure Connections (for Data Sources for Profiles)
  6. (Optional) Configure Data Sources from Connections (for Profiles)
  7. (Optional) Configure Profile schemas from Data Sources.
  8. Configure and deploy the actions that the Skill runs.
  9. Define Skills for the Agent (build image, push to repository, and invoke).
  10. Create the Agent record.
  11. Add Skills (and skill properties) to the Agent.
  12. Configure service messages for the Agent (Add inputs/outputs).
  13. Connect service message as inputs and outputs to Skills.
  14. Map data types for the message flows from the Service messages through the Skills.
  15. Invoke the Agent in the development cluster (to train and test).
  16. When the Agent is ready to be deployed generate an Agent snapshot. (Agent Snapshots are versions of an Agent that contain the entire bill-of-materials required to run that agent.)
  17. Use FabricOps to deploy Agent Snapshot in another Fabric cluster (e.g. production or model office).

SENSA Fabric Agent Building interface​

SENSA Fabric Console's Agent Composer provides a web-based visual AI composition workbench that provides a higher level programming model for the implementation of Enterprise AI systems. It is designed to make composition, orchestration, testing, and deployment of AI agents as simple as possible for application developers.

For details about developing Agents go to the Build Agents section of the docs.