@sisu-ai/rag-core

Reusable backend-agnostic RAG mechanics for Sisu.

Exports

• chunkers: characterChunker, sentenceChunker, paragraphChunker, getChunker
• direct helpers: prepareRagRecords, storeRagContent, retrieveRagContext
• types for embeddings providers, chunking, and retrieval/storage results

Philosophy

@sisu-ai/rag-core is where reusable RAG mechanics live.

• It is not a middleware package.
• It is not a model-facing tool package.
• It is not tied to any one vector backend.

Its job is to turn text and embeddings into vector-store operations in a small, composable, backend-agnostic way.

Package role

Use @sisu-ai/rag-core when you need RAG mechanics outside tool-calling, such as startup seeding or developer-controlled ingestion.

• @sisu-ai/tool-rag wraps this package for model-facing tool calls
• @sisu-ai/vector-core provides vector contracts
• @sisu-ai/vector-chroma provides a Chroma-backed VectorStore
• @sisu-ai/vector-vectra provides a local file-backed VectorStore

What It Owns

• chunking strategies
• text-to-record preparation
• embeddings orchestration
• direct store/query helpers over a VectorStore
• retrieval result shaping into compact citation-ready results

What It Does Not Own

• model-facing tool schemas or descriptions → @sisu-ai/tool-rag
• middleware prompt injection → @sisu-ai/mw-rag
• vector backend SDK code → @sisu-ai/vector-chroma, @sisu-ai/vector-vectra, or another vector-* package

Typical Flow

1. Prepare records without writing yet

import { prepareRagRecords } from '@sisu-ai/rag-core';

const prepared = await prepareRagRecords({
  content: 'Long-form content goes here.',
  embeddings,
  chunkingStrategy: 'sentences',
  chunkSize: 400,
  overlap: 1,
});

2. Store content directly

import { storeRagContent } from '@sisu-ai/rag-core';

await storeRagContent({
  content: 'Important context to persist.',
  embeddings,
  vectorStore,
  chunkingStrategy: 'sentences',
});

3. Retrieve context directly

import { retrieveRagContext } from '@sisu-ai/rag-core';

const result = await retrieveRagContext({
  queryText: 'What does the user prefer?',
  embeddings,
  vectorStore,
  topK: 4,
});

How It Fits With Other Packages

• Use @sisu-ai/rag-core directly in app code for ingestion and reusable retrieval logic.
• Use @sisu-ai/tool-rag when the model should call storage/retrieval itself.
• Use @sisu-ai/mw-rag when the app controls embeddings and retrieval explicitly in middleware.

The same embeddings provider and vectorStore can be shared across all three.

Contributing

We build Sisu in the open. Contributions welcome.

Contributing Guide · Report a Bug · Request a Feature · Code of Conduct

Star on GitHub if Sisu helps you build better agents.

Quiet, determined, relentlessly useful.

Apache 2.0 License