---
title: Add RAG
description: Attach AgentsKit retrieval so answers cite your corpus.
---

# Add RAG

AgentsKit Chat does **not** reimplement RAG. Wire [AgentsKit retrievers](https://www.agentskit.io/docs/data/rag) into the same `chat` config you pass to `defineChat`.

## Two planes

<Mermaid
  chart={
    'flowchart TD\n' +
      '  Q[User question] --> Det{Deterministic local}\n' +
      '  Det -->|exact match| Cite[Cited local answer]\n' +
      '  Det -->|miss| Ask[Ask backend + RAG]\n' +
      '  Ask --> Ret[AgentsKit Retriever]\n' +
      '  Ret --> Gen[AgentsKit provider]\n' +
      '  Gen --> Out[Cited streaming answer]'
  }
/>

## Pattern

```ts
import { defineChat } from '@agentskit/chat'
import type { AdapterFactory } from '@agentskit/core'

export const createDocsChat = (adapter: AdapterFactory) =>
  defineChat({
    id: 'docs',
    chat: {
      adapter,
      systemPrompt: 'Cite sources. Prefer links over guesses.',
      // attach AgentsKit memory / tools / retriever the same way as any AgentsKit app
    },
  })
```

## Grounded Ask (site assistants)

When you need semantic retrieval with citations:

1. Host [`createAskServiceHandler`](/docs/backend)
2. Inject retriever + generator adapters
3. Client uses `createAskAdapter` + optional deterministic artifact

```ts
import { createAskAdapter, createDeterministicAnswerAdapter, defineChat } from '@agentskit/chat'

const ask = createAskAdapter({ endpoint: '/api/ask', corpus: 'public', persona: 'guide' })
const adapter = createDeterministicAnswerAdapter({
  artifact: verifiedKnowledge,
  fallbackMode: 'backend',
  fallback: ask,
})

export const chat = defineChat({ id: 'docs', chat: { adapter } })
```

## AgentsKit docs

- [RAG package](https://www.agentskit.io/docs/data/rag)
- [Providers / adapters](https://www.agentskit.io/docs/data/providers)
- [Memory](https://www.agentskit.io/docs/data/memory)

## Related

- [Backend](/docs/backend)
- [Connect backend](/docs/guides/connect-backend)
- [Examples](/docs/examples)