ragPipeline()

Creates a reactive retrieve-augment-generate pipeline.

Signature

function ragPipeline(opts: RagPipelineOptions): RagPipelineResult

Parameters

Parameter	Type	Description
opts	RagPipelineOptions	Pipeline configuration (query store, search sources, LLM store).

Returns

RagPipelineResult — reactive context, docs, generating, error stores + destroy().

Basic Usage

import { state } from 'callbag-recharge';
import { ragPipeline, docIndex, fromLLM } from 'callbag-recharge/ai';

const query = state('');
const docs = docIndex({ db: '/docs-index.db' });
const llm = fromLLM({ provider: 'ollama', model: 'llama4' });

const rag = ragPipeline({ query, docSearch: docs, llm });

// Trigger retrieval + generation
query.set('How do I use derived stores?');
// rag.generating.get() → true
// rag.context.get() → "SEARCH RESULTS:\n[1] ..."
// llm.get() → accumulating response...

Options / Behavior Details

• Reactive query: Set opts.query to a non-empty string to trigger retrieval + generation.
• Async semantic search: If semanticSearch is provided and loaded, waits for embedding result before generating. Uses latestAsync to cancel stale in-flight searches on rapid query changes.
• Context assembly: SYSTEM PROMPT → SUMMARY → USER CONTEXT → SEARCH RESULTS.
• Cleanup: destroy() cancels in-flight searches, unsubscribes from query, and tears down derived stores. Does not destroy passed-in stores (llm, docSearch, etc.).