Rasa is an open-source machine learning framework for automating text- and voice-based conversations, primarily aimed at building chatbots and voice assistants. According to its repository description, it provides natural language understanding (NLU) and dialogue management capabilities, enabling developers to create contextual assistants that can handle a large number of multi-turn exchanges, that is, dialogue systems capable of handling layered, continuous conversation scenarios.
Regarding channel integration, Rasa can connect to Facebook Messenger, Google Hangouts, Microsoft Bot Framework, and users' own custom conversation channels; the repository description also mentions that it can connect to platforms such as Slack and Facebook. This allows the same set of dialogue logic to be deployed to multiple messaging and voice entry points.
It should be specially noted that the project's current status: the repository clearly states "Rasa Open Source is currently in maintenance mode." The official statement says that the future direction for building AI agents with Rasa is Hello Rasa and CALM.
Regarding the subsequent direction, the material introduces Hello Rasa: it is a new interactive playground for prototyping AI agents, which, through the CALM (Conversational AI with Language Models) engine, combines the fluency of large language models with the reliability of business logic. Its features include: no installation or configuration required; after opening the playground, select a template (banking, telecom, support) and start building in the browser; no NLU training required, that is, it has moved beyond the intents approach, with the large language model responsible for dialogue understanding and developers defining business flows; a built-in copilot, where a dedicated AI assistant helps generate code, debug flows, and instantly extend the agent; and it is production-ready, that is, not just a toy, and when ready to scale, the agent can be exported to the Rasa Platform. CALM's positioning is to combine the flexibility of large language models with strict business logic: the large language model understands the user, while code enforces the rules. Flows replace rigid dialogue trees with a way of describing logical steps (such as collecting payments or making transfers). The Inspector allows viewing the real-time decision-making process, avoiding a black box. The material also provides a link to start building for free (Hello Rasa) as well as a vendor-neutral Agent Engineering community for discussing architecture, memory, orchestration, and security, and for influencing the Hello Rasa roadmap and getting early access to features.
Regarding classic Rasa Open Source (Legacy), the material explains that its documentation and installation instructions apply to this traditional framework. When installing the development version, Rasa uses Poetry for packaging and dependency management: first install Poetry according to the official guide, then build from source by running make install; macOS users who encounter compiler issues can try export SYSTEM_VERSION_COMPAT=1 before installation. Rasa follows Semantic Versioning: major versions are incompatible API changes, minor versions are backward-compatible features, and patch versions are backward-compatible bug fixes. The full release cadence and maintenance policy are described in its product release and maintenance policy. The project is licensed under the Apache License 2.0, Copyright 2022 Rasa Technologies GmbH. The repository also provides Contribution Guidelines and a Code of Conduct, and welcomes contributions.
Regarding applicable scenarios, the material points to conversational AI projects that need to build contextual dialogue assistants, chatbots, or voice assistants and want to connect to multiple messaging channels, or that are premised on using their own servers and data control. Limitations and notes: Rasa Open Source has entered maintenance mode, and new features and the future direction have shifted to Hello Rasa and CALM; specific capability boundaries, performance data, charges and pricing, supported service levels, and so on are not yet provided in the material.
