Rasa Review 2026: CALM, Pricing, and the Best Alternatives

Rasa has a reputation as the serious, developer-first option for conversational AI. Banks, telcos, and healthcare teams run it because it gives them something most platforms don't: full control over the model, the data, and where everything lives. That control is real, and for some teams it's the whole reason to choose Rasa.

It also comes with a bill, and not just the dollar kind. Rasa asks for engineering time, infrastructure to run, and a team comfortable owning all of it. Before you commit, it's worth understanding what Rasa actually is in 2026, what it costs, and where a managed platform makes more sense. Here's the honest version.

What's New in Rasa (2026)

If your mental model of Rasa is "the open-source Python framework with intent files," it's out of date. Three things have changed:

• Rasa moved to CALM. CALM (Conversational AI with Language Models) is Rasa's LLM-native approach to dialogue. Instead of classifying every message into a pre-defined intent, the language model handles understanding while your business logic stays in code. Rasa positions this as a high-trust, low-hallucination way to build with LLMs.
• There's a no-code option now. The platform splits into Rasa Pro (the pro-code framework and infrastructure) and Rasa Studio (a no-code UI built on top of it). The old "Rasa has no menus or dashboards" line no longer holds.
• The free tier was repackaged. The Rasa Developer Edition replaces the old "Rasa Open Source" framing and ships with CALM, free to run locally and in production within set conversation limits.

What Is Rasa?

Rasa is a framework for building, testing, and deploying conversational AI agents (chatbots and voice assistants) that you host and control yourself.

What sets Rasa apart is where the intelligence lives. With CALM, the language model interprets what the user means, but the steps your agent can actually take are defined by you in code. That separation is the point: it keeps the agent from improvising its way into a compliance problem, which is why regulated industries gravitate to it.

Key Features of Rasa

• CALM dialogue: The LLM-native layer that reads user intent in context, then executes flows you've defined. It works with frontier models like GPT-4 and with smaller fine-tuned models (down to Llama 8B) when you need low latency for voice.
• Natural language understanding: Rasa's NLU heritage is still here for teams that want classic intent and entity handling alongside or instead of CALM.
• Rasa Studio: A no-code interface for building and editing agents without writing Python, aimed at the non-developers on a conversational AI team.
• Self-hosting and open foundations: You can run Rasa in your own environment and inspect or extend it down to the source. For teams with strict data-residency rules, that's the headline feature.
• Integrations: Rasa connects to channels and services from WhatsApp to custom voice stacks, so it slots into the tools you already run.

How Much Does Rasa Cost?

Rasa's pricing is tiered, and the free tier is more generous than most people expect. Here's the current structure:

The takeaway for budgeting: the framework is free to start, but the moment you need real volume or support, you're in five-figure-annual territory and a sales conversation. That's normal for self-hosted enterprise software. Just don't mistake "open-source" for "free at scale."

Building a Rasa Chatbot

Building on Rasa means working in code. You install Rasa, define your flows and domain, and iterate in a normal Python toolchain (VS Code or any editor you like), running and testing the agent locally before you deploy. With CALM, you're describing the steps the agent can take and letting the model handle the conversation around them, rather than hand-writing training phrases for every intent.

This is genuinely powerful, and it's genuinely work. You're responsible for hosting, scaling, monitoring, and maintaining the stack. If you have engineers who want that control, Rasa rewards them. If you don't, that ongoing cost is the thing to weigh honestly before you start.

{{blue-cta}}

The Best Alternatives to Rasa

Rasa isn't the only way to build a serious conversational AI agent, and for many teams it's not the fastest. A few platforms worth comparing:

1. IBM Watson Assistant: IBM's enterprise conversational AI, with pre-built industry templates, multi-language support, and reporting. Another heavy, enterprise-leaning option.
2. Amazon Lex: Amazon's NLU and speech-recognition service. It integrates tightly with the rest of AWS, which is a plus if you're already there and a lock-in risk if you're not.
3. Voiceflow: A no-code-to-pro-code platform for building chat and voice agents, with model choice, native voice, and built-in evaluation and observability. More on the head-to-head below.

For a wider field, see our roundups of the best AI agent builders and the best AI chatbots, and our look at the leading open-source chatbot platforms if self-hosting is a hard requirement. For another self-host-vs-managed comparison, Google Dialogflow and Botpress make useful reference points.

Voiceflow vs. Rasa

Both platforms build capable agents. The real difference is who has to do the work, and how much you can see once the agent is live.

Who builds it. Rasa is developer-first; even Rasa Studio sits on top of infrastructure your team hosts and maintains. Voiceflow is no-code-to-pro-code: CX teams build on a visual canvas, and developers extend through the API and SDK. You get a faster path to production without standing up and running the stack yourself.

Which models you run. Rasa's CALM works with the models you wire in. Voiceflow is model-agnostic by design: pick OpenAI, Anthropic, Google, Bedrock, or Groq per agent, or bring your own. If avoiding model lock-in matters, that flexibility is built in.

How the agent behaves. Voiceflow splits agent logic into two primitives that compose. Workflows are deterministic SOPs for tasks that must go right every time, like refunds or compliance checks. Playbooks give the agent a goal and room to reason. A workflow can hand off to a playbook mid-conversation and back again, so the agent stays adaptable without becoming unpredictable. It's the same control Rasa gives you in code, available without code.

What you can see. Good agents, like good employees, need good management. Voiceflow includes a Knowledge Base, Evaluations, and an observability suite so you can trace every conversation, define what "good" looks like, and test changes in staging before they ship, all managed rather than self-hosted. It's SOC 2 Type 2 compliant with PII masking for teams with compliance requirements, and security and data-handling questions are worth weighing carefully when you evaluate any agent platform for enterprise use.

The honest version: if you must self-host and you have the engineering to own it, Rasa is a strong choice. If you want production speed, native voice, and management tooling without running infrastructure, that's where Voiceflow fits. Voiceflow's customers include Turo, StubHub International, Sanlam Studios, and Trilogy. To see how it handles your use case, book a demo.

{{blue-cta}}

Frequently Asked Questions

What is a Rasa chatbot?

A Rasa chatbot is a conversational AI agent built on Rasa's framework, which you host and control yourself. In 2026, Rasa builds these agents with CALM, an approach where a language model handles understanding while the actions the agent can take stay defined in code.

Is Rasa free to use?

Partly. The Rasa Developer Edition is free and includes CALM, covering up to 1,000 conversations per month (100 per month for internal employee agents). Beyond that volume, you move to the Growth tier, which starts around $35,000 per year, or a custom Enterprise plan.

What's the difference between Rasa and ChatGPT?

ChatGPT is a consumer assistant and an API you call. Rasa is a framework for building your own agents, where you control the flows, host the system, and choose which model powers it. You might even use a model like GPT-4 inside a Rasa agent through CALM. They solve different problems.

Is Rasa open source?

Rasa has open-source roots, and the Developer Edition lets you run and inspect it yourself. The commercial Growth and Enterprise tiers (Rasa Pro and Rasa Studio) add support, scale, and features on top. So the foundation is open, but the full platform is a paid product.

How much does Rasa cost?

The Developer Edition is free within its conversation limits. The Growth tier starts around $35,000 per year for higher volume and support, and Enterprise pricing is custom. Both paid tiers are quote-based through Rasa's sales team.