Documentation
Foundations of Wayfinder AI's architecture, data integration and operation.
Overview
Wayfinder AI is an on-premise platform for demand intelligence. It reads existing company data (read-only), produces daily demand forecasts and translates them into concrete action for purchasing, production and sales.
This documentation is a first scaffold and will grow to cover operations, integration and model details.
Architecture (on-premise)
The entire Wayfinder AI stack runs inside your network — on your server or VM. There is no cloud requirement; internet access is not needed to operate it.
Components: data connectors (read-only), a feature and forecasting pipeline, steering/rule logic, a web interface and optionally a locally hosted language model for plain-text reports.
Data integration
Sources are connected step by step: ERP/inventory, production, sales and external signals (season, holidays, weather, events). Integration via SQL, CSV or API — always read-only.
Data that is legally or technically restricted is never forced. Data quality is made transparent before forecasts are used productively.
Forecast & accuracy
Forecasts are produced per article, customer and location and refreshed daily. Accuracy is shown openly (e.g. MAPE), including handling of outliers and seasonality.
No black box: results are traceable and measured against actuals.
Security & permissions
Access to source systems is read-only; source systems are never modified. A role and rights concept ensures users only see what they are allowed to. Accesses and runs are auditable.
No data leaking into third-party training pools, no dependency on external model APIs, data exportable at any time.
12-week timeline
1–2: Target & data · 3–5: Connection · 6–8: Forecast · 9–10: Steering · 11–12: Pilot.
Result: a first productive decision basis on real data with a clear measure of success.
Integration questions?
Write tosupport@wayfinder-ai.de.