RAG

Retrieval-augmented generation: chunk documents, embed them, store the vectors, and retrieve the most relevant context for a query.

Chunk & embed

rag.ts

import { MDocument } from '@redtuma/rag'

const doc = MDocument.fromText(longText)
const chunks = doc.chunk({ strategy: 'recursive', size: 512, overlap: 64 })

const embedded = await embed(chunks, embedder)

Store & query

Vector stores share a common VectorStore interface (upsert / query / delete). Query by an embedded vector to get the top matches.

await vector.upsert(embedded)

const matches = await vector.query({ queryVector, topK: 5 })

As an agent tool

Expose retrieval as a tool so the agent can pull context on demand.

const agent = new Agent({
  id: 'researcher',
  instructions: 'Answer using retrieved context only.',
  model: 'anthropic/claude-opus-4-8',
  tools: { search: ragTool },
})

← Model routing MCP →