Rasa: agent building features and setup

Overview

Rasa combines language model capabilities with developer-defined workflows and rules. It suits cases where a conversation needs to lead through a business process, such as checking a request, collecting details, or invoking a tool. The focus is on control across many conversations, rather than answering a single question. A project like this also requires system integrations, conversation design, and testing, so it is suited to a team with development and maintenance capabilities.

Features

• Combines language models with business logic in a conversation workflow.
• Connects tools and systems to perform actions and use knowledge.
• Tools for building, deploying, and monitoring agent performance across channels.

Business uses

• An internal help agent that guides an employee through a defined request process.
• A service workflow that collects details and performs an action after appropriate verification.

Getting started

• Map one conversation workflow, including changing the subject, correcting details, and requesting a human representative.
• Connect actions to test systems and define the data required for each action.
• Test Hebrew conversations with varied phrasing before a broad release.

What to check

• Check which capabilities are included in the edition and license that fit the project, rather than relying on an example from another product.
• Keeping business logic does not remove the need to verify a user's identity before showing personal information or performing an action.