Knowledge base + dashboard + MCP
Your archive, made queryable.
We turn years of documents, contracts, budgets and email into a knowledge base classified by offering, with a KPI dashboard and an MCP server: management queries it in plain language, from Claude or any AI assistant, one token per person.
The problem
The company knows more than it knows it knows: the knowledge sits in files, not in the people who are still there.
Every board question costs someone a day: search, reconcile, rebuild the spreadsheet.
The offerings that actually pay are not the ones you think: without classifying the archive you never find out.
What we deliver
Classified knowledge base
We index the archive (full-text + semantic search) and classify every document by offering, client, period and relevance with an LLM benchmarked on a gold set: you know how reliable it is before you rely on it.
Offering map
What you actually sell, to whom, at what margin and with what recurrence. The basis for pricing, positioning and lead generation.
KPI dashboard
Revenue, margin, pipeline, cash and headcount in one view reconciled to the cent with the sources. Periodic cuts, not diverging spreadsheets.
MCP server for management
One secure endpoint, one token per member, drill-down tools: “how much did we bill on offering X in the last 12 months?” is answered in ten seconds, with the source.
How we work
Discovery (2 weeks)
Source audit, classified sample, preliminary offering map, build plan with scope and risks. You leave with a decision document, not a demo.
Build (4–8 weeks)
Index, full classification, dashboard, MCP server in production with named access. Automated tests and documented reconciliation.
Run (monthly)
New cuts, new sources, tool maintenance, one monthly review with management.
A case
For a Milan-based B2B services company we indexed over 6,000 documents and 50,000 text blocks, classified the offering, published a treasury and forecast dashboard and put an MCP server with 26 tools into production. The board received its tokens and started asking questions the same day.
Frequently asked questions
Where does the data live?
On your infrastructure or on a dedicated server in Europe; no training data shared with model providers.
Do we need a modern ERP?
No. We work with exports, spreadsheets, PDFs and email. Reconciliation is part of the job.
Who can query it?
Whoever holds a token. One per person, revocable, with a question log.
Does it work with our AI assistant?
With any MCP client: Claude, ChatGPT, Cursor and others.
Bring a question your archive cannot answer today.
In 30 minutes we tell you whether and how it can.
Prefer to write first?