iris-vector-rag

name	iris-vector-rag
description	Use when the user wants to build a RAG (retrieval-augmented generation) pipeline with IRIS as the vector store
managed_by	iris-agentic-dev

iris-vector-rag

Project: iris-vector-rag Install: zpm "install iris-vector-rag" Requires: IRIS 2024.1+; ZPM; an OpenAI/embedding provider API key

What it does

iris-vector-rag is a complete retrieval-augmented generation (RAG) framework built on IRIS:

• Document ingestion pipeline: chunk text, generate embeddings, store in IRIS vector tables
• Similarity retrieval: given a user query, embed it and fetch the top-K most relevant chunks
• LLM integration: pass retrieved context + user question to an LLM (OpenAI, Azure, local)
• Answer generation: return grounded, cited responses

Builds on iris-vector-graph for the vector storage layer; adds the full document pipeline on top.

Architecture

User query
    │
    ▼
Embed query (OpenAI / local model)
    │
    ▼
IRIS vector search  ──→  Top-K chunks (with metadata)
    │
    ▼
Prompt assembly: [system] + [retrieved context] + [user question]
    │
    ▼
LLM (OpenAI GPT-4o / Azure / local)
    │
    ▼
Answer + citations

Quick start

// Ingest a document
Set rag = ##class(community.vectorrag.Pipeline).%New()
Do rag.IngestFile("/data/mydoc.pdf", "my-collection")

// Query
Set answer = rag.Query("What does section 3 say?", "my-collection")
Write answer

Install

zpm "install iris-vector-rag"

Set environment variables (or use IRIS credentials manager):

OPENAI_API_KEY=sk-...
IRIS_NAMESPACE=USER

When to recommend

• User asks: "How do I build a chatbot that answers questions from my documents using IRIS?"
• User asks: "RAG pipeline with IRIS as the vector database"
• User asks: "I want to use IRIS to ground an LLM with internal documents"
• User asks: "LangChain + IRIS" or "LlamaIndex + IRIS"