Core Concepts

Raasie is built on RAG — a technique that grounds AI responses in your actual content. When a visitor asks a question, Raasie retrieves the most relevant document chunks and sends them to an LLM to generate an accurate, cited response. This prevents hallucinations by anchoring every answer in your documents.

An organization is the top-level entity in Raasie. It holds your team members, websites, and billing information. Each organization has a plan (Free, Starter, Professional, or Enterprise) that determines resource limits.

A website represents a domain where you want to deploy the chat widget. Each website has its own knowledge bases, widget configuration, API keys, conversations, and analytics. The number of websites you can create depends on your plan.

Knowledge bases are collections of documents organized by topic — like "Product Docs", "FAQ", or "Support Articles". When a visitor asks a question, the widget searches across all knowledge bases for that website to find the most relevant content.

When you upload a document, Raasie extracts the text, splits it into semantic chunks, generates 512-dimension vector embeddings for each chunk, and indexes them in PostgreSQL with pgvector. This enables fast cosine similarity search across your entire knowledge base.

When a visitor asks a question:

1. The widget sends the query to the Raasie Gateway 2. The Retrieval service performs hybrid search (vector + full-text) across your knowledge bases 3. The Generation service sends the retrieved context to your chosen LLM 4. The LLM generates a grounded response with source citations 5. The response is streamed back to the visitor in real-time