Scope agents to assigned stores, territories, customers, and product categories.
The org structure becomes the access structure — a store’s agent works that store, a category manager’s agent works that category.
Control agents across customers, stores, suppliers, payments, inventory, and workforce systems.
Retail agents operate across ecommerce, stores, customer support, merchandising, inventory, loss prevention, and vendor management. A single workflow can touch customer data, payments, pricing, and physical operations.
Example day. The stores, roles and grants are the ones you set.
The org structure becomes the access structure — a store’s agent works that store, a category manager’s agent works that category.
Each is its own action in the grant, and it also has to be in the acting associate’s own permissions — the smaller of the two sets wins.
Each sensitive domain is its own scope, so a service workflow never doubles as access to payment or workforce data.
Committing money is not an action the agent holds, so a merchandiser or manager is the one who does it.
Case-scoped access keeps every conversation inside the account it belongs to.
When margin or shrink questions surface, the full sequence behind each action is one query away.
Explore the context attached to every request.
retail-agent has a defined purpose and a bounded set of tools.
The request acts for store manager, using that person’s permissions.
The policy applies to the requested action within Retail platform, before the tool executes.
Service agents resolve issues within the customer’s own account, and only with the actions the associate they assist already holds.
Return agents process the returns they were granted; anything outside that set is refused before it reaches the system, so it lands with a supervisor.
Onboarding agents gather documents and draft records; activation is not an action they hold, so procurement does it.
Store agents work their own location’s data — schedules, tasks, inventory — without cross-banner reach.
Rebalancing agents read stock levels and propose transfers; committing one is not in their grant, so a planner does it.
Merchandising agents draft price and assortment changes within their category; publishing them is a separate action they are not given.
Employee-facing assistants answer from each worker’s own records and published policy, nothing broader.
Yes. Fabriq sits between your agents and the order, payment, inventory, and support systems you already run — no re-platforming. Actions keep flowing through the same tools, each one checked against the agent’s grant and the acting user’s own permissions first.
Payment, loyalty, employee, and customer data are separate scopes. A service agent resolving an order issue works with order and account context without raw payment access, and every read of sensitive data is recorded.
Access follows identity — grants resolve org → team → member, and revoking a session cuts a person and their agents off immediately.
Yes. Which refund, discount and purchasing actions are reachable resolves per acting user and location, so one agent fleet serves every store while each associate’s agent operates within exactly that associate’s authority.
Yes — lookups and drafts stay instant while price changes, bulk edits and refunds are left out of the grant, so they never appear in the tool list. Every verdict is logged.
Through one hub — prebuilt connectors plus OpenAPI import for internal services and Postgres behind read-only guardrails, all logged on the way through.