AI Operations Hub
The operations hub where a pro-audio e-commerce company runs its main processes, plans, and analyzes its data, with agentic workflows and machine learning built in.
- Client
- Koala Music
- Industry
- Pro-audio e-commerce
- Role
- Product, architecture, and development

Capabilities
What we built
The company's operations hub: orders, products, stock, invoices, contacts and replenishment are handled in one place, in sync with the ERP.
Integrates the ERP, the e-commerce store, sales channels such as marketplaces, and other internal company tools.
Planning intelligence, data and analytics for the business, in the same hub.
Agentic workflows and machine learning models support parts of the operation.
An MCP server that connects Claude, ChatGPT or any MCP client to the hub, with the same capabilities as the UI.
Every capability works two ways: in the web app and through MCP, so people and AI assistants can do the same things.
A data-analysis agent that answers questions like "how many orders are stuck in picking?" with clear, structured answers.
Secure by design: users sign in with OAuth, and every question respects the person's role and permissions.
- An in-app AI assistant, so the team can ask questions without leaving the hub.
Context
The product
Koala Music sells pro-audio gear online and runs its operations on the Bling ERP. It needed one hub, connected to its ERP, e-commerce store, marketplaces and internal tools, for its main processes (orders, products, stock, invoices, contacts and replenishment) plus the planning intelligence, data and analytics side of the business, and a safe way to ask questions about it in plain language from the AI assistants the team already uses.
How it's built
Technology
A TypeScript platform with a NestJS API and a Next.js web app on PostgreSQL, integrated with the Bling ERP, the e-commerce store, marketplaces and other internal tools. Agentic workflows (Mastra, with Anthropic and OpenAI models) and machine learning models support parts of the operation, the planning and analytics features, and the data-analysis agent. The MCP server exposes the business data as tools for AI assistants, with OAuth sign-in and per-role permissions. Every capability is exposed both in the UI and through the MCP server. Everything runs in Docker on a VPS.
Stack
- MCP
- Mastra
- Vercel AI SDK
- Anthropic / OpenAI
- Machine learning
- OAuth 2.0
- NestJS
- PostgreSQL
- Next.js
- Bling ERP
- Docker
Outcome
Results
Main operations, planning, data and analytics in one hub, connected to the ERP, e-commerce, marketplaces and internal tools.
Agentic workflows and machine learning models supporting parts of the operation.
The team queries the business straight from the AI tools they already use.
Live in production.
Gallery
Screens

Replenishment planning: turnover, days of cover and reorder suggestions per SKU (demo data). 
Orders from every sales channel in one place, with status counters and filters (demo data). 
Connecting Claude to the hub: a personal MCP token with ready-to-paste setup (demo data; token masked). 
Stock per product and warehouse (demo data). 
Finance: accounts payable with overdue tracking (demo data). 
Integrations: sync routines configurable from the UI (demo data). 
Product catalog with stock alerts (demo data).
Want something like this?
Tell us about your use case. We will tell you honestly whether an agent is the right tool and what it would take.