@sisu-ai/tool-rag

Backend-agnostic RAG tools for agents.

Exports

• createRetrieveTool
• createStoreTool
• createRagTools

Type exports are re-exposed for convenience, but reusable chunking and ingestion mechanics live in @sisu-ai/rag-core.

Philosophy

@sisu-ai/tool-rag is the model-facing layer of the RAG stack.

• It is where tool schemas, names, and descriptions live.
• It stays thin by delegating the actual chunking, embeddings orchestration, and vector-store work to @sisu-ai/rag-core.
• It works with any VectorStore implementation that satisfies @sisu-ai/vector-core.

Composition

@sisu-ai/tool-rag expects:

• embeddings provider (embed(input[]) => vectors[])
• vectorStore implementation (upsert, query)

Use backend adapters such as @sisu-ai/vector-chroma or @sisu-ai/vector-vectra to provide vectorStore.

For startup seeding, chunking helpers, and developer-controlled ingestion, use @sisu-ai/rag-core directly.

How It Resolves Dependencies

You can provide dependencies in two ways:

• directly in tool options
• indirectly through ctx.deps

That means the same tool factory works for:

• explicitly wired apps
• middleware/orchestration flows that inject dependencies into tool context

Typical Usage

import { createRagTools } from '@sisu-ai/tool-rag';
import { createChromaVectorStore } from '@sisu-ai/vector-chroma';

const vectorStore = createChromaVectorStore({ namespace: 'docs' });

const ragTools = createRagTools({
  embeddings,
  vectorStore,
  store: {
    chunkingStrategy: 'sentences',
    chunkSize: 400,
    overlap: 1,
  },
});

This gives the model:

• retrieveContext for query-text retrieval
• storeContext for communication-derived persistence

When To Use This Package

• You want the model to decide when to retrieve context.
• You want the model to store user-provided information for later retrieval.
• You want backend-agnostic RAG tools rather than provider-specific ones.

When Not To Use This Package

• You want app-controlled ingestion or retrieval without tool-calling; use @sisu-ai/rag-core.
• You want deterministic middleware composition; use @sisu-ai/mw-rag.

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