All case studies

Warehouse logistics

FoxMag WMS with AI

A warehouse system with a picker app, an office panel, and an AI assistant that knows the warehouse layout and, through MCP tools, suggests where to put stock and the shortest way to pick an order.

warehouse operating modes
3
WMS tools for AI agents
MCP

Client

Logistics and warehouse companies - from warehouses with full location addressing to businesses that only need stock tracking.

Problem

Order picking done from memory, stock put wherever there was space, and inventory in spreadsheets. Knowledge about the warehouse lived in a few people's heads.

Scope of work

Design and build of a WMS: mobile app, web panel, API, shared model package, identity, and an AI assistant. MVP scope for a single-warehouse pilot with a UAT protocol and a go-live runbook.

Solution

Pickers work in a Flutter app (receiving, picking, stocktake, scanning) with offline support, the office in a panel with orders, CSV import, transfers, label printing, and roles. A Dart (Shelf) API exposes endpoints for critical paths plus /health and /metrics, with data and identity in PostgreSQL on Supabase. The AI assistant (Python / FastAPI) first classifies intent, then hands it to an agent - slotting, picking, forecasting, stocktake, safety - which acts through MCP tools on the real API. The prompt receives a "world model": the warehouse ontology, the company operating profile, and the current operations snapshot.

Key features

  • Receiving, picking, transfers, and stocktakes
  • Mobile picker app with offline mode
  • Office panel: orders, CSV import, printing, roles, and permissions
  • Three operating profiles: full WMS, locations, stock only
  • AI agents: slotting, picking, forecasting, safety
  • MCP tools and a warehouse world model for the assistant
  • Local (Ollama) or cloud model, with an AI off switch

Business outcome

MVP ready for a single-warehouse pilot: a UAT protocol for receiving, picking, and stocktake, a go-live runbook with rollback, and a policy for the AI assistant's role. Agent evaluation and benchmarks run on a separate test suite.

Technologies

FlutterDartShelfSupabasePostgreSQLPythonFastAPIMCPRAGOllamaDocker