RAG Mesh Router

TECHNICAL DEMONSTRATION

Why several RAGs might serve you better than a single one

Many organizations now have teams building their own “shadow AI” solutions. These may address immediate needs, but they are rarely designed as governed, domain-owned specialists that can cooperate across the organization.

Most RAG implementations still rely on a single blended index. This page demonstrates another approach: a RAG Mesh—independent, domain-owned systems that retain their own knowledge, retrieval strategies, and access controls. A Router determines which specialist should retrieve and answer, while preserving clear responsibility for who maintains each knowledge domain.

Illustrative enterprise mesh

One Router. Three owned knowledge domains.

DOMAIN 01

Marketing RAG

Customers · campaigns · brand

Owned by Marketing

ROUTER RAG Router

Selects specialist(s)

No shared corpus

DOMAIN 02

Manufacturing RAG

Processes · quality · maintenance

Owned by Operations

DOMAIN 03

Finance RAG

Forecasts · controls · reporting

Owned by Finance

Each domain keeps its own knowledge · retrieval · access · maintenance

The question behind the demo is simple: should knowledge this different really share the same index, retrieval strategy, access model, and maintenance process? To make the contrast as clear as possible, we selected three deliberately different RAGs from our own inventory: Mark Twain’s literature, NASA Space Talks transcripts, and—just for the fun of it—an isolated RAG about RAG itself.

How it works Ask → Route → Retrieve → Answer · expand for background

The router knows where to ask—not everything the specialists know.

Each specialist owns its corpus, retrieval strategy, permissions, and answer boundaries. The Mesh supplies discovery, policy-aware routing, and a common way to ask. When you press Route question, the page walks four inspectable stages — it does not jump straight to a chatbot reply.

01Ask
Your question, unchanged. Starters show single-domain vs cross-domain routes.
02Route
Rule-based selection: Selected or Not needed, with an explicit reason.
03Retrieve
Passages each selected specialist pulled — evidence before the answer.
04Answer
Equal Tutorial and Tech: plain lead → how to think about it → literature, cited.

Federated by design

Knowledge stays with the team that owns it — no mega-vector pile on the Router.

Authorization travels with the query

Discoverable does not mean accessible; routing can skip a node on purpose.

Evidence before synthesis

Route and retrieve stay inspectable; the Mesh does not merge specialists into one fake voice.

Mark Twain checking

Literary questions over Twain’s complete works — grounded chat, not free invention.

NASA Space Talks checking

Mission-ops from Mercury–Apollo transcripts — who said what, with attribution.

Rag of Rags checking

RAG research itself — chunking, retrieval, evaluation. Mesh is the UI.

Ask across the different domains. The router finds it way, either he consults one, two or all when needed. Select a conversation starter below, or construct your own query.