What Rasa is
Rasa is a mature conversational AI framework: intent classification and entity extraction through a trained NLU pipeline, plus dialogue management through rules, stories, and policies. It is Python-native, self-hostable, and has been the default answer for high-volume scripted customer service assistants for years. Rasa Pro adds commercial tiers, including LLM-driven dialogue.
Shared problem, different generation of solution
EDDI and Rasa both come out of the conversational AI world, and EDDI has been in continuous development as a conversational platform since 2006. The difference is where the intelligence sits. Rasa puts it in a model you train on your own labeled data, which gives precise, repeatable control. EDDI puts it in orchestrated LLMs governed by declarative configuration, trading some determinism for reach and speed of change.
Dialogue, tooling, and the cost of change
| Dimension | Rasa | EDDI |
|---|---|---|
| Understanding | Trained NLU: intent classification and entity extraction from labeled data | LLM comprehension plus intent-based agent discovery and routing |
| Changing behavior | Update training data, retrain, evaluate, redeploy | Edit JSON configuration; no retraining and no redeployment |
| Determinism | High. Rules and policies make flows repeatable and testable | Lower by nature, recovered through behavior rules and approval gates |
| Language | Python; custom actions are Python code | JVM runtime; agent logic is configuration, tools are declared |
| Tool ecosystem | A custom action server you build and host | 84 MCP tools, built-in agent tools, plus any OpenAPI endpoint |
| Multi-agent | One assistant per deployment is the common pattern | Coordinator-based orchestration and 7 group discussion styles |
| Licensing | Community edition plus commercially licensed Rasa Pro tiers; verify current terms | Apache 2.0, OSI approved |
| Compliance | Self-hosting supports data residency; frameworks implemented per deployment | Audit ledger, data subject rights API, and 15+ frameworks addressed in the platform |
Details about the other tool were checked against its official documentation and repository in October 2026. These projects change quickly, so confirm current behavior with the vendor before you decide.
Choosing between them
Where Rasa is the better choice
- You already have labeled conversational data and an ML team who can maintain an NLU pipeline.
- Regulatory or brand constraints require a deterministic assistant with an auditable decision path that does not involve a language model.
- Very high volume over a narrow intent set, where per-turn LLM cost would dominate.
- You need fine-grained control over classification confidence thresholds and fallback behavior.
Where EDDI is the better choice
- The assistant must answer from documents and systems that change faster than a retraining cycle.
- Business users need to change behavior without an ML engineer, a training run, or a deployment.
- The problem needs several specialized agents to collaborate rather than one flat intent tree.
- You want LLM flexibility, but a compliance officer requires approval gates and signed logs before it goes live.
- You want to avoid lock-in at the model layer: 19 providers, or a local model through Ollama or Jlama.
Moving an assistant from Rasa
The mapping is conceptual rather than mechanical, because the two locate intelligence in different places. Expect to translate intent structure into routing and prompts rather than to convert files.
| Rasa | EDDI |
|---|---|
| Intents and training examples | Intent-based agent discovery; the labeled data becomes evaluation material rather than training input |
| Entities and slots | Structured extraction plus persistent user memory |
| Stories and rules | Behavior rules and pipeline configuration |
| Custom action server | MCP tools or httpCall tool definitions |
| Responses and domain file | Prompt snippets, reusable and versioned |
| Fallback policy | Human-in-the-loop escalation with a timeout policy |
Frequently asked
Is EDDI just another LLM wrapper replacing trained NLU?
EDDI has been developed as a conversational platform since 2006, well before the current LLM generation, and it kept the parts that mattered: intent-based routing, managed conversation state, and pipeline architecture. LLMs replaced the comprehension layer, not the orchestration around it.
We need deterministic answers for compliance. Does that rule out EDDI?
Not necessarily, but the trade should be explicit. LLM output is not deterministic. EDDI addresses this with behavior rules that constrain responses, human approval gates on consequential actions, and a signed audit ledger recording what was decided and why. If your requirement is that no language model participates in the decision at all, Rasa is the better answer.
Can we run both during a transition?
Yes. A common pattern routes known high-volume intents to the existing Rasa assistant while EDDI handles open-ended queries and system integration, then shifts traffic as confidence grows. EDDI can call a Rasa endpoint through httpCall.
Other comparisons
- EDDI vs. LangChain and LangGraph
- EDDI vs. Microsoft Agent Framework (AutoGen and Semantic Kernel)
- EDDI vs. CrewAI
- EDDI vs. Dify
- EDDI vs. n8n
- EDDI vs. Flowise
- All alternatives: builders, libraries, and cloud platforms
Ready for production?
Get managed deployment, SLA-backed support, and hands-on help from the team that builds EDDI.
Talk to us about enterprise deployment