Rasa

3.6.21 · active · verified Thu Apr 16

Rasa is an active open-source machine learning framework (version 3.6.21) designed to automate text and voice-based conversations. It provides tools for Natural Language Understanding (NLU) and dialogue management, enabling developers to create contextual chatbots and voice assistants that can integrate with various platforms like Slack and Facebook. The project maintains a regular release cadence with frequent updates.

Common errors

```
ModuleNotFoundError: No module named 'rasa'
```
cause Rasa is not installed in the current Python environment, or the virtual environment is not activated, or the system is using a different Python interpreter than where Rasa was installed.

fix Ensure your virtual environment is activated (`source .venv/bin/activate` on Linux/macOS, `.venv\Scripts\activate` on Windows). If Rasa is not installed, run `pip install rasa` within the activated environment.
```
Couldn't connect to the server at 'http://localhost:5055/webhook'. Is the server running?
```
cause The Rasa action server, which hosts custom actions, is not running or is not accessible at the configured address (default: `http://localhost:5055/webhook`). This is common in Docker setups or when running components separately.

fix In a separate terminal, navigate to your Rasa project directory and run `rasa run actions`. If using Docker, ensure your `docker-compose.yml` correctly exposes and links the action server, and that the `endpoints.yml` inside the Rasa container points to the correct service name/IP within the Docker network (e.g., `action_endpoint: url: "http://action_server:5055/webhook"`).
```
Error when trying to connect to 'mongod' tracker store. Using 'InMemoryTrackerStore'' instead. The causing error was: localhost:21476: [Errno 111] Connection refused
```
cause Rasa failed to connect to the configured external tracker store (e.g., MongoDB, Redis). This usually means the database server is not running, is inaccessible, or there are incorrect connection details/credentials in `endpoints.yml`.

fix Verify that your database server (e.g., MongoDB or Redis) is running and accessible from where Rasa is launched. Check the `endpoints.yml` file for correct `url`, `port`, `db`, `username`, and `password` settings. Ensure necessary network ports are open if connecting remotely or across Docker containers.

Warnings

breaking Rasa 3.6.21 introduced a model breaking change by replacing `pickle` and `joblib` with safer alternatives like `json`, `safetensors`, and `skops` for component serialization. Models trained with older versions will not be compatible.
Fix: Retrain your Rasa models after upgrading to Rasa 3.6.21 or newer. If you have custom components that override `persist` or `load` methods, update your code to use the new serialization mechanisms.
gotcha Rasa 3.x has specific Python version requirements. The `rasa` PyPI package currently requires Python <3.11, >=3.8. Using incompatible Python versions can lead to `ModuleNotFoundError` or other installation issues.
Fix: Always install Rasa in a virtual environment with a supported Python version (e.g., `python3.10 -m venv .venv && source .venv/bin/activate`). Check the official documentation for the exact compatible Python versions for your Rasa release.
deprecated Rasa X, the UI for Rasa, does not support Rasa Open Source versions 3.x and above. Attempting to use `rasa x` with Rasa 3.x will result in an error.
Fix: If you need to use Rasa X, you must downgrade your Rasa Open Source installation to version 2.8.x. For Rasa 3.x and newer, consider alternative UI tools or develop custom interfaces.
breaking Rasa 3.0 introduced a major architecture revamp, transitioning from a sequential training pipeline to a graph architecture. This significantly changed how NLU and policy components interact and can affect custom components, particularly the `process` method signature for NLU components (now accepting `List[Message]` instead of `Message`).
Fix: Review and update any custom NLU components or policies to align with the new graph architecture and method signatures, especially for `process` methods that handle incoming messages.

Install

pip install rasa Basic Installation
pip install 'rasa[full]' Installation with all optional ML dependencies

Imports

Agent
```
from rasa.core.agent import Agent
```
Action
```
from rasa_sdk import Action
```

CollectingDispatcher

from rasa_sdk.executor import CollectingDispatcher

TrackerStore
wrong:
```
from rasa.core.tracker_store import MongoTrackerStore
```
correct:
```
from rasa.core.tracker_store import TrackerStore
```
For custom tracker stores, inherit from base `TrackerStore`. Specific implementations like `MongoTrackerStore` are part of the internal API or configured via `endpoints.yml`.
Interpreter
wrong:
```
from rasa_nlu.model import Interpreter
```
correct:
```
from rasa.nlu.model import Interpreter
```
The `rasa_nlu` package was deprecated; NLU functionalities are now integrated into the main `rasa` package.

Quickstart

The standard way to get started with Rasa is through its command-line interface. This quickstart initializes a new project, trains a basic model, and then allows interactive conversation with the bot in the shell. Custom actions require running a separate action server.

import os

# Ensure you are in an empty directory for this quickstart
# In a real project, replace 'tutorial' with your desired template or omit for default
# os.system('rasa init --template tutorial') 
# os.system('rasa train')
# os.system('rasa shell')

print("To start a new Rasa project, open your terminal and run:")
print("1. rasa init --template tutorial")
print("2. cd <your-project-name>")
print("3. rasa train")
print("4. rasa shell")
print("For custom actions, run 'rasa run actions' in a separate terminal after step 3.")

view raw JSON →