Skip to content

Rasa Architecture Components

Rasa is an open-source framework for building contextual assistants. This diagram shows how a user message moves through its layers: from a messaging channel, through the NLU pipeline that understands the message, into the Core that decides what to do, out to custom actions that touch real systems, and back as a response. Rasa X tooling feeds new training data back into the models. Hover any component for details.

Interactive Demo

Run MicroSim Fullscreen

To embed this MicroSim in your own page, use the following iframe:

1
<iframe src="main.html" width="100%" height="802" scrolling="no"></iframe>

Overview

The diagram is grouped into color-coded layers:

  • User Input (blue) - messaging channels such as Slack, Teams, and a web widget.
  • Rasa NLU Pipeline (green) - tokenizer, featurizer (word embeddings), intent classifier, and entity extractor.
  • Rasa Core (orange) - the Tracker Store (conversation history), the Dialog Policy (ML or rule-based), and the Action Server.
  • Custom Actions (purple) - database connectors, external API clients, and business logic functions.
  • Rasa X Tooling (gold) - conversation review, training-data annotation, and a performance dashboard.
  • External Systems (gray) - the databases and APIs reached by custom actions.

Solid arrows show message flow. The Tracker Store and Dialog Policy share a bidirectional state-tracking link. Dashed arrows show Rasa X feeding training data back into the NLU and Core, and custom actions making API calls to external systems.

Lesson Plan

  • Trace one message. Have students follow a customer message from the web widget all the way to a database lookup and back to the user.
  • Separate NLU from Core. Ask students to explain why Rasa splits "what the user said" (NLU) from "what to do next" (Core).
  • Close the loop. Discuss how Rasa X turns real conversations into new training data, and why that feedback loop improves the bot over time.
  • Custom actions. Identify which tasks require a custom action versus a simple response template.

References