Vimaponto: talk to your operation instead of navigating it.

Vimaponto runs its industrial operation on its planning system, VMP Plan, connected to the Primavera ERP. All the information is there — orders, production plans, statuses, deadlines. The hard part is reaching it: to know what's overdue, or to change a plan, someone has to navigate the software, open screens and run queries.

The idea was to put a layer of conversation on top of what already exists: anyone asks in natural language — “what's overdue?”, “what's the status of this plan?” — and gets an answer grounded in real data. And not just answers: the agent also triggers actions in the system itself.

In an industrial context, an AI agent that makes up answers is worthless — worse, it's risky. The challenge wasn't “add a chatbot”, it was wiring the AI to the real systems reliably: always using true data as the source, knowing what it can and can't do, and being able to execute actions without breaking anything. Without that, it's just talk.

We thought the product through end to end — architecture, integrations, design and the structural questions — before writing the first answer.

At the centre is an agent (FastAPI and LangGraph, with Gemini) that treats the operational software as the source of truth. Instead of wiring it field by field, we used the MCP protocol: VMP Plan exposes its entities and actions as tools the agent discovers and uses — read records, understand statuses, and create, update or execute. The rule is strict: never invent data; if the information doesn't come from the system, say so.

Around it, what makes this usable day to day: persistent per-conversation memory, a sense of real time (overdue, deadlines, “due today”), reading of uploaded documents, and Microsoft 365 integration for flows like schedule changes. Plus the link to the Primavera ERP, where the operation runs.

For the people using it, a conversational interface (Next.js) with charts and diagrams generated by the agent itself — so the answer isn't only text.

Today there's a working demo: you ask the operation in natural language and get answers grounded in VMP Plan's real data, with the agent also able to trigger actions on the system. The product — architecture, integrations and design — is built to grow from pilot to production.

If your operation lives inside an ERP and a planning system, the value isn't in replacing them — it's in letting your people talk to them and act, with the guarantee that the answer comes from real data. That's what we built.