Rasa Integrates RAG with 'Enterprise Search Policy'
Conversational AI platform Rasa has introduced an “Enterprise Search Policy” to integrate RAG capabilities directly into enterprise assistants. The policy supports both direct knowledge base lookups and LLM-powered contextual responses. This is part of Rasa's new CALM (Conversational AI with Language Models) approach, which blends structured Q&A with generative answers to meet modern enterprise expectations.
- The Enterprise Search Policy integrates with multiple vector stores out-of-the-box, including local options like Faiss and managed instances like Milvus and Qdrant, allowing developers to connect to existing RAG pipelines. It also supports custom information retrievers for proprietary search systems or advanced filtering logic. - Rasa's CALM framework is designed as a hybrid system that separates dialogue understanding from execution; it uses an LLM to interpret a user's intent but relies on predefined, deterministic "Flows" to control the assistant's actions, which is intended to prevent hallucinations. - The framework is engineered to be model-agnostic, supporting state-of-the-art models like GPT-4, but is also optimized to function with smaller, fine-tuned models such as Llama 8B to improve response times and reduce operational costs. - For enterprise use, Rasa Pro offers tiered pricing; while there is a free Developer Edition with a cap of 1,000 monthly external conversations, the "Growth" tier for teams starts at $35,000 annually. - Rasa's move comes as Gartner identified RAG as a core capability for enterprise generative AI in 2024, signaling a broader market shift from experimental to production-critical RAG architectures to ground LLMs in verifiable data. - The policy includes a `check_relevancy` feature, which allows the assistant to assess if a user's query has a relevant answer in the knowledge base and abstain from responding if it doesn't, preventing inaccurate guesses. - In the conversational AI market, Rasa's open-source, developer-focused approach competes with platforms like Google's Dialogflow, which offers a more visual state-machine-based logic, and Botpress, which also provides an open-source option but with a greater emphasis on a visual, low-code interface.
