@sisu-ai/vector-vectra

Vectra vector-store adapter for Sisu RAG composition.

Exports

• createVectraVectorStore

The created adapter exposes upsert, query, and delete, and is intended to be injected into backend-agnostic RAG tools such as @sisu-ai/tool-rag.

For reusable chunking, record preparation, and app-side seeding flows, pair this adapter with @sisu-ai/rag-core.

Philosophy

@sisu-ai/vector-vectra owns only Vectra-specific translation.

• It implements the shared VectorStore contract from @sisu-ai/vector-core.
• It does not own chunking, prompt shaping, or tool schemas.
• It uses Vectra LocalIndex, which fits Sisu's existing split where @sisu-ai/rag-core already owns chunking and embeddings orchestration.

Setup

npm i @sisu-ai/vector-vectra vectra

Usage

import { openAIEmbeddings } from '@sisu-ai/adapter-openai';
import { storeRagContent } from '@sisu-ai/rag-core';
import { createVectraVectorStore } from '@sisu-ai/vector-vectra';

const embeddings = openAIEmbeddings({ model: 'text-embedding-3-small' });
const vectorStore = createVectraVectorStore({
  folderPath: '.vectra',
  namespace: 'travel',
  indexedMetadataFields: ['docId', 'source'],
});

await storeRagContent({
  content: 'Malmö fika notes go here.',
  source: 'seed',
  metadata: { docId: 'malmo-guide' },
  embeddings,
  vectorStore,
  namespace: 'travel',
});

Namespaces

Vectra has no built-in namespace primitive, so this adapter maps each namespace to its own local folder under folderPath.

• base folderPath: .vectra
• namespace travel: .vectra/travel
• namespace docs: .vectra/docs

Queries against a namespace that has not been written yet return an empty match set.

Metadata

• Scalar metadata values are stored directly.
• Non-scalar metadata values are JSON-stringified before persistence.
• Filterable fields should be listed in indexedMetadataFields when creating the adapter.

This keeps Vectra-specific metadata constraints inside the adapter package instead of leaking into rag-core or tool-rag.

How It Fits With The RAG Stack

• @sisu-ai/vector-core defines the shared storage contract
• @sisu-ai/vector-vectra implements that contract with file-backed local indexes
• @sisu-ai/rag-core handles chunking and direct store/retrieve flows
• @sisu-ai/tool-rag exposes model-facing retrieval/storage tools
• @sisu-ai/mw-rag composes deterministic middleware-driven retrieval over any VectorStore

Contributing

We build Sisu in the open. Contributions welcome.

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

• @sisu-ai/rag-core

• @sisu-ai/vector-core
• @sisu-ai/vector-chroma
• @sisu-ai/vector-vectra

Star on GitHub if Sisu helps you build better agents.

Quiet, determined, relentlessly useful.

Apache 2.0 License