What Is Rasa AI? Pricing, CALM and Alternatives (2026 Guide)

Rasa is a Berlin-based developer platform for building enterprise AI agents, founded in 2016, whose CALM framework pairs an LLM for understanding language with deterministic, developer-written business logic. Its Free Developer Edition allows one bot and up to 1,000 external conversations a month in production; everything above that is custom-quoted Enterprise pricing (rasa.com/pricing, checked 24 September 2026).

We build AI agents for customer service ourselves, so we read Rasa the way a technical evaluator would: what you actually get on the free tier, what changes once you need a paid plan, and where the open-framework pitch differs from a closed platform.

What is Rasa?

Rasa was founded in 2016 and is headquartered in Berlin, with a San Francisco office, built on the open-source NLU and dialog-management work the founding team started that year. Alan Nichol, co-founder, remains CTO; current leadership includes CEO Melissa Gordon, Chief Product Officer Björn Minkmar, and Chief Revenue Officer Vikas Bhambri. Investors named on Rasa's about page include Andreessen Horowitz, PayPal Ventures, Accel, and Basis Set Ventures, though Rasa doesn't disclose a specific funding total or round amounts publicly.

Rasa positions itself squarely at technical teams in regulated industries, finance, healthcare, insurance, telecom, and government, where "full access to prompts, policies, and codebase" and on-premises or offline deployment options matter as much as the conversational AI itself. Named customers on Rasa's homepage include Autodesk, BNP Paribas, Groupe IMA, Swisscom, Providence, Orange, Albert Heijn, N26, ERGO, and T-Mobile. Rasa was named a Strong Performer in Forrester's 2026 Wave for Conversational AI Platforms for Customer Service, per Rasa's own citation of the report.

How does Rasa's CALM framework work?

The core technical argument is CALM (Conversational AI with Language Models), and it's a genuine architectural choice, not just a marketing label. Rasa's own framing: most conversational AI platforms either rely on an LLM for everything, improvising every response, or hide their decision logic in an opaque "composite" system nobody can fully audit. CALM instead separates language understanding (handled by an LLM) from business logic (handled by deterministic, developer-defined flows), so an agent can hold a natural conversation while still guaranteeing it never skips a required compliance step or promises something the business logic doesn't allow.

That comparison matters concretely for cost and reliability: a legacy hybrid platform calls the LLM constantly and locks into fixed classifiers that handle one task at a time, struggling when a user changes topics mid-conversation. CALM avoids unnecessary LLM calls specifically to keep latency and cost down, while still letting the LLM interpret free-form language and help phrase responses. Rasa reports, in its own figures that we could not independently audit, a 50% cost reduction across more than 50 enterprise deployments and a 59% goal-completion rate per conversation using this approach.

What features does Rasa include?

• CALM framework: separates LLM-driven language understanding from deterministic business logic.
• Open framework: full access to prompts, policies, and the underlying codebase, with no vendor lock-in.
• Multi-agent orchestration: coordinates domain-specific agents while maintaining conversational state.
• Voice/IVR support: real-time speech processing with turn-taking and barge-in handling.
• Flexible deployment: on-premises, private cloud, or fully offline environments, a genuine differentiator for regulated industries.
• Real-time observability: built-in monitoring, evaluation, and performance tracking.

How much does Rasa cost?

Rasa publishes a genuinely usable free tier: the Free Developer Edition gives one bot per company, up to 1,000 external conversations a month (or 100 internal conversations), usable locally or in production, with community forum support. That's a real production allowance, well beyond a sandbox, which is unusual in this category.

Beyond that, Enterprise is custom-quoted, requiring a sales conversation, and adds full platform access, premium support with 24/7/365 response times, a dedicated Customer Success Manager and engineer, and self-managed or managed deployment options. The same page also describes a Business plan that bundles Rasa Pro with Rasa Studio, the no-code side (an AI assistant flow builder and a content and response management system). It lists no price for Business either, and it isn't one of the two headline cards in the screenshot above. Budget for a sales conversation once the 1,000-conversation ceiling is a real constraint.

Giving away a usable free tier is its own incentive: a developer who builds a working bot on Rasa's own infrastructure and workflow is the easiest possible lead for the sales team once that bot needs to scale past what the free tier allows.

What are Rasa's pros and cons?

Pros

• A genuinely usable free production tier (1,000 conversations/month), not just a trial.
• Full framework transparency: prompts, policies, and codebase are yours to inspect and modify, with no vendor lock-in.
• On-premises and fully offline deployment options, a real requirement for some regulated industries.
• CALM's separation of language from logic is a defensible architectural answer to LLM hallucination risk in a business-critical flow.

Cons

• No published pricing beyond the free tier; Enterprise requires a full sales cycle.
• The framework's flexibility comes with real technical setup: it's a developer platform first, with a point-and-click builder as an add-on rather than the default experience.
• A very thin independent review sample, 11 reviews on G2 and 6 on Capterra, checked 21 September 2026, so most satisfaction figures in circulation are Rasa's own.
• Neither the Business plan nor Enterprise has a published price, so you can't compare total cost with other platforms before talking to sales.

What do Rasa users say in reviews?

When this guide was first researched, G2 and Gartner Peer Insights both blocked our access and the Capterra page we tried resolved to an unrelated product. Reading them in a real browser on 21 September 2026 got further: Rasa holds 4.0 out of 5 from 11 reviews on G2 and 4.7 out of 5 from 6 reviews on Capterra. Seventeen reviews across both sites is too small a sample to generalise from, but it is real independent signal, and G2's 4.0 sits below the 4.4 Rasa cites on its own homepage. What's available is Rasa's own citation of a 4.4-star customer satisfaction figure on its homepage, alongside the Forrester Wave 2026 recognition and the 50%-cost-reduction, 59%-goal-completion statistics discussed above, all vendor-reported figures we could not independently source.

How we researched this: we checked Rasa's homepage, pricing, platform, and CALM pages directly on 2026-09-17, plus rasa.com/about for company background. G2 and Gartner Peer Insights both returned blocked responses to a direct fetch, and this session's WebSearch budget was used up before an independent review source could be found, so third-party review data was a gap in the original version of this guide. On 21 September 2026 we filled it by reading G2 and Capterra in a real browser; both samples are small and are quoted above with the date checked, and Gartner Peer Insights still refused access. On 24 September 2026 we re-read Rasa's homepage, pricing and about pages: the free-tier limits, leadership, Forrester citation and vendor statistics were unchanged, and the pricing page now clearly describes the Business plan.

Who is Rasa best for?

Rasa fits engineering teams at regulated enterprises, finance, healthcare, insurance, telecom, government, that need full control over prompts and decision logic, potentially on-premises or fully offline, and that have the technical capacity to build on an open framework rather than configure a closed platform. The free tier's real 1,000-conversation production allowance also makes it a reasonable starting point for a technical team validating an approach before an Enterprise conversation.

Rasa fits teams with engineering capacity more than it fits a non-technical team wanting a managed, point-and-click agent with someone else handling the infrastructure. Rasa's core pitch, full framework access and deployment flexibility, is precisely the thing a team without engineering resources doesn't need and would pay an ongoing setup cost to use well.

How does Rasa compare with Kore.ai and Macha?

Rasa and Kore.ai sit at genuinely different points on the build-versus-buy spectrum: Rasa gives a technical team a framework to build and fully control, while Kore.ai sells a more managed enterprise platform through a sales-led process. The choice between them usually comes down to whether an organization has (or wants) the engineering capacity to own its conversational AI stack directly.

Macha sits outside this comparison entirely: instead of a framework for building a conversational AI agent from scratch, it's an AI agent layer that runs on top of a help desk a team already uses (Zendesk, Freshdesk, Gorgias, Front, HubSpot or Intercom), and it's done for you: the Macha team analyzes your past tickets, builds the knowledge base and the agent, and runs it in safe mode until the drafts are right, rather than a developer team building on an open framework. A team that wants to own its own conversational AI stack fits Rasa; a team that wants a working agent on its existing help desk without a build project fits Macha. Macha costs about $0.40 a ticket from $299/month for 750 tickets, which includes setup and a dedicated success manager who handles your changes. See Macha's pricing or start with $50 of free usage against real tickets.

The other obvious comparison for a team that wants to own its stack is Botpress, which sits between Rasa's framework and a closed platform, and our Kore.ai guide covers the enterprise end of the same market.

FAQ

What is Rasa? Rasa is an open, developer-first platform for building enterprise conversational AI agents, founded in 2016, built around its CALM framework that separates LLM-driven language understanding from deterministic business logic.

How much does Rasa cost? The Free Developer Edition allows one bot per company with up to 1,000 external conversations a month, usable in production, at no cost. The Business plan (Rasa Pro plus Rasa Studio) and Enterprise have no published prices and require a sales conversation.

Is Rasa open source? Rasa is built on open-source foundations, and its framework gives full access to prompts, policies, and the underlying codebase, with no vendor lock-in, though its paid tiers are commercial products layered on top.

What is CALM? CALM (Conversational AI with Language Models) is Rasa's architecture for separating language understanding (handled by an LLM) from business logic (handled by deterministic flows), aimed at keeping an agent reliable and auditable rather than fully improvised.

What do real users say about Rasa? Independent review data is thin but no longer absent: 4.0 out of 5 from 11 reviews on G2 and 4.7 out of 5 from 6 on Capterra, both checked 21 September 2026, while Gartner Peer Insights still blocks access. Rasa's own site cites a 4.4-star customer satisfaction figure and a Forrester Wave 2026 Strong Performer recognition.

Who is Rasa best for? Engineering teams at regulated enterprises (finance, healthcare, insurance, telecom, government) that need full control over their conversational AI stack, potentially with on-premises or offline deployment requirements.

Is Macha an alternative to Rasa? Not directly. Rasa is an open framework for building conversational AI agents from scratch; Macha is an AI agent layer that runs on top of a help desk a team already uses (Zendesk, Freshdesk, Gorgias, Front, HubSpot or Intercom), built and managed by the Macha team. A team that wants to own its stack fits Rasa; a team that wants a working agent without a build project fits Macha.

Sources: Rasa, Rasa pricing, Rasa Platform, Rasa CALM, Rasa about. G2 and Gartner Peer Insights both blocked direct fetches; G2 and Capterra figures were read in a browser on 21 September 2026.