RAG / document Q&A

Chat interface grounded in your own documents — PDFs, Markdown files, knowledge bases.

A RAG app lets users ask questions against a private corpus. The agent retrieves the most relevant chunks, injects them as context, and cites the source. The standard shape is:

document ingestion (chunking + embedding)
vector retrieval at query time
chat UI with context injection
optional source display

#Who it's for

Internal knowledge bases (runbooks, HR policies, product docs)
Customer-facing help (support knowledge, product manuals)
Research tools over large document sets (legal, financial, academic)

#Typical stack

npm install @agentskit/react @agentskit/adapters @agentskit/rag @agentskit/memory

#Recommended package mix

Layer	Package	Why it matters
UI	`@agentskit/react`	Chat surface with streaming and source display
Provider	`@agentskit/adapters`	OpenAI, Anthropic, Gemini — swap without rewriting
RAG	`@agentskit/rag`	Chunking, embedding, retrieval in one call
Vector store	`@agentskit/memory`	In-memory for dev, LanceDB for production

#Architecture

documents → rag.ingest() → vector store
user query → rag.search() → top-K chunks → LLM context window → answer

#60-second snippet

import { createRAG } from '@agentskit/rag'
import { inMemoryStore } from '@agentskit/memory/vector'
import { useChat } from '@agentskit/react'
import { openai } from '@agentskit/adapters/openai'

// 1. Build the RAG retriever once (module level)
const rag = createRAG({
  store: inMemoryStore(),
  embed: async (text) => {
    // call your embedding API here
    const res = await fetch('/api/embed', {
      method: 'POST',
      body: JSON.stringify({ text }),
    })
    return (await res.json()).embedding as number[]
  },
  topK: 5,
})

// 2. Ingest documents at startup (idempotent — safe to re-run)
await rag.ingest([
  { id: 'policy-v2', content: policyText, source: 'policy.pdf' },
  { id: 'faq-2024', content: faqText, source: 'faq.md' },
])

const adapter = openai({ model: 'gpt-4o-mini' })

// 3. Wire the retriever into useChat
export function DocChat() {
  const chat = useChat({ adapter, retriever: rag })
  // …render as usual
}

Tip

Swap inMemoryStore() for lanceStore({ path: './vectors.lance' }) when your corpus grows past a few thousand chunks. The retriever contract is identical.

Pitfall

Create the rag instance once at module level, not inside the component. createRAG allocates the store on construction — calling it on each render resets the vector index.

RAG in 15 lines — minimal standalone snippet
Tools + memory together — adding state to RAG

@agentskit/rag — createRAG, loaders, chunkers, rerankers
@agentskit/memory — inMemoryStore, lanceStore, redisStore
@agentskit/adapters — provider adapters

Explore nearby

✎ Edit this page on GitHub·Found a problem? Open an issue →·How to contribute →