embedding-rag

Simple RAG (Retrieval-Augmented Generation) pipeline using the built-in native embedding executor -- keyword-indexed SQLite store, pure Go, zero external dependencies.

Usage

kdeps run examples/embedding-rag/workflow.yaml --dev

Index a document:

curl -X POST http://localhost:16403/index \
  -H "Content-Type: application/json" \
  -d '{"text": "Go is a statically typed, compiled programming language designed at Google.", "collection": "docs"}'

Search indexed documents:

curl -X POST http://localhost:16403/search \
  -H "Content-Type: application/json" \
  -d '{"query": "compiled language", "collection": "docs"}'

How it works

1. index -- run.embedding with operation: upsert stores the text in a SQLite keyword index
2. search -- run.embedding with operation: search retrieves the top-5 matching documents
3. response -- returns the results and match count

Operations

operation	Description
index	Add document to index
upsert	Add or update document
search	Keyword search, returns ranked results
delete	Remove document by text

Structure

embedding-rag/
├── workflow.yaml
└── resources/
    ├── index.yaml    # run.embedding operation: upsert
    ├── search.yaml   # run.embedding operation: search
    └── response.yaml # API response