Commerce for agents — the Catalog Brain
A structured, agent-ready data layer in development. It translates your store's products, relationships, and policies into a common vocabulary machines can read — the foundation every FaStart AI capability is designed to read from, and our long-term moat in agentic commerce.
This is being built
Early AccessThe Catalog Brain is the data foundation of our agentic roadmap — in early access, not a shipped feature. The layers and protocol flow below describe what we're building and the order we're building it in; live functionality switches on once the backend is connected. Join the waitlist to shape it with us.
The Catalog Brain (early access) is FaStart's structured, agent-ready data layer: a product graph, normalized attributes, a knowledge base, and semantic embeddings. It aims to give shopping agents a common protocol vocabulary to discover and transact against your catalog, powering AI search, answer-engine optimization (AEO), and the Agentic Commerce Gateway. Clean, connected commerce data is the long-term moat — and the thing every other AI capability reads from.
Agent-ready data layers
From a connected product graph up to semantic embeddings — each layer is a building block of a common vocabulary agents can read, and makes the next AI capability possible.
Product graph
It aims to model products, variants, collections, and the relationships between them as one connected structure — so links like “bought with”, “alternative to”, “parent category” become machine-readable.
Normalized attributes
It aims to map attributes that every seller writes differently — color, size, material — onto a consistent schema. “Lacivert / navy / dark blue” resolve to one concept, making filtering and comparison reliable.
Knowledge base
It aims to hold store knowledge — size charts, care instructions, shipping and return policies — in structured form, so assistants and agents read from the source instead of guessing.
Semantic embeddings
It plans to represent products and queries as semantic embeddings to support intent-based search like “summery linen shirt”. This layer is on our roadmap.
What the Catalog Brain powers
One clean data layer, designed to feed every AI surface and agent workflow above it.
AI search
It aims to let in-store search move past keywords to understand intent — finding what the shopper actually meant.
AEO / answer engines
Structured product data forms the foundation that lets answer engines like ChatGPT and Perplexity cite your store accurately.
Agentic Commerce Gateway
It aims to provide the clean, agent-ready data layer that external shopping agents need to read your catalog and transact.
Walk through an agent flow, step by step
How a shopping agent discovers the catalog, creates a cart, updates it, and completes the order — see exactly what data is exchanged at each step. This is an illustrative preview, not a live endpoint.
The agent learns the store's capabilities and version.
GET /.well-known/ucp# capability discovery
Accept: application/json{
"ucp.version": "early-access",
"capabilities": {
"discovery": true,
"checkout": true,
"fulfillment": true
},
"data_layers": [
"product_graph",
"normalized_attributes",
"knowledge_base"
]
}Agents and humans, working together
Every agent flow is designed to escalate cleanly wherever human approval is needed. A dynamic handoff based on cart and checkout context aims to give every transaction a path to completion — automation that works without taking control away from you.
- Handoff to a human at sensitive steps (payment, address)
- Dynamic handoff based on cart and checkout context
- Merchant of record stays with the seller — FaStart is not in the money flow
Why data is the moat
As shopping shifts toward AI assistants and autonomous agents, the stores that win will be the ones whose catalog data is clean, connected, and machine-readable. A flashy chatbot on top of messy data still gives wrong answers. We're investing in the layer underneath first — because every AI capability above it is only as good as the data it reads.
- A common protocol vocabulary removes the integration meeting for agents
- A single source of truth — assistants and agents read instead of guess
- Clean data is reused as new discovery surfaces appear
Build your store on the right foundation
The Catalog Brain is in early access, but the platform underneath is live today. Join the waitlist to help shape the protocol layer with us.