Rasa's open-source chatbot framework enters maintenance mode

The conversation

Rasa is not a single vendor's hosted product but software a team installs and runs itself. The project's own GitHub README, retrieved 16 September 2026, describes it plainly: Rasa is an open source machine learning framework for automating text and voice-based conversations, released under the Apache License, Version 2.0. The same README lists messaging surfaces a self-hosted Rasa assistant has historically connected to, including Facebook Messenger, Slack, Microsoft Bot Framework and Telegram. Rasa's own documentation landing page now markets a newer commercial layer, the Rasa Platform, split into a pro-code tier called Rasa and a no-code tier called Studio.

What the documents show

The README is explicit that what most people mean by Rasa, the classic open-source framework with separate natural-language-understanding and dialogue-management components, is now in a different position than it once was. It states that Rasa Open Source is currently in maintenance mode, and names Hello Rasa and an engine called CALM, for Conversational AI with Language Models, as the future of building AI agents with Rasa. Rasa's own words about CALM are direct: the company says it has moved beyond intents, so a language model handles dialogue understanding while a developer defines the business flows. The classic architecture documentation itself, reached through a legacy documentation domain, still describes the older model in terms of NLU training data and stories, material used to train understanding of a message and the conversational paths a policy component chooses between.

The system boundary

Because Rasa is software a team runs rather than a hosted platform, the boundary of what the framework does for a business is set by what that team chooses to deploy and operate, not by a vendor's servers. The legacy documentation's own guidance on deploying assistants, covering how to deploy the core assistant and a separate action server, makes that operational split explicit: understanding a message and deciding a dialogue policy is one deployable component, and any custom business logic a team writes is a second, separately hosted piece. Nothing in the classic framework calls out to an external company's infrastructure by default; whatever a bot does beyond a scripted reply is code the operating team wrote and must keep running.

Where it fails

The open question this history raises for a builder is not whether Rasa's classic model works, but which Rasa a given tutorial, blog post or job listing is actually describing. A README stating that its own core product is in maintenance mode is telling operators that new investment is going elsewhere, even while the existing framework remains installable and documented.

• Is a given reference to Rasa describing the classic NLU-and-dialogue-management framework, or the newer CALM-based Hello Rasa product?
• Who on the team owns patching, scaling and monitoring the self-hosted service, since no vendor does it automatically?
• If the classic framework stays in maintenance mode, what is the plan for security and dependency updates over the deployment's lifetime?

This is an editorial checklist, not a claim about Rasa's technical merits either way: the sources establish that a transition is under way, not how it will end.