Artiplane frequently asked questions

Artiplane is the AI operating layer for fashion and luxury: it connects enterprise systems, unifies them in a fashion semantic model, runs governed agents and Digital Workers on top, and delivers the result in a role-based Worker App.

BI shows the past. Artiplane proposes the action, simulates it, routes it to whoever has authority and measures the outcome after execution.

Six: Systems, Users, AI layer, Semantic layer, Digital Worker and Digital App (the Worker Apps).

An agent performs a task across one or more systems. A Digital Worker is the alter ego of a role: it coordinates several agents and workflows, proposes activities and stays subject to human approval.

It is the operational application of a role: buying, pricing, planning, supply chain, clienteling, customer service, factory. It is built and renamed in the Control Plane and opens full screen.

SAP, PLM, PIM, CRM, POS, WMS, commerce, data platforms and automation tools such as n8n: the demo landscape counts more than 35 applications.

No. Every write is a registered action with its target system, input contract, approval threshold and execution record. Agents can request one; they cannot invent one.

With a field-by-field mapping onto the canonical model, versioned through draft, test and publish, always with visible lineage.

Yes: weather by market, competitor assortments and price moves, trend signals, all mapped onto the company's semantic categories.

Anthropic Claude, OpenAI, Google Gemini, native runtimes and n8n or SAP BTP workflows. The model is chosen per agent and can be switched off or replaced from the AI Hub.

Each RAG base is a governed object with sources, refresh, per-area permissions and measured cost: answers always cite the documents used.

Numbers come from the semantic layer, not from the model; the model explains and proposes. Guardrails, control steps and human approval close the loop.

Yes: the AI Hub shows last-30-days cost, requests and tokens per model, agent, RAG and runtime, with drill-down to a single agent.

Yes: every reasoning step names the model that runs it, and agents imported from market LLMs always show their provenance.

Every logged interaction carries an SHA-256 compliance hash; content storage is opt-in and retention is configurable.

With risk classes per use case, mandatory human oversight where required, documented model provenance and configuration audit.

Isolation level is a configuration choice: standard, EU-hosted only or on-premise, with the models allowed for each level.

No. Actions shown in the public experience are not sent to enterprise systems. Access is provided to explore behaviours, controls and operational flows.

Two: Admin, which configures the platform, and User, which operates. Admin does not approve business decisions; the user does not change configuration.

Per Worker App you define authorized users with restrictions: geographies, product categories, read-only or write, budget and operating window.

The Digital Worker proposes them or the user creates them; they can be one-off or recurring, they carry dependencies and close with an outcome, attachments and a downloadable output.

Like in SAP: each change becomes a logged Change Request, with environment and version visible, transported to production in one click.

You start from one process and the systems feeding it: connection, mapping, one Digital Worker and its Worker App are configuration work, not development.

No. You need people who know the processes: the Control Plane is designed for IT and function owners, with no code to write.

From the access page, with the code we provide: you enter as Admin or as User, in the language you prefer.