How TaxCompass answers — and why you can trust it
TaxCompass doesn't answer from memory. Every reply is built from the actual Italian rules and shows you the source behind each sentence, so you (and your commercialista) can verify it.
How an answer is built
It reads the primary sources
The knowledge base is built from official texts — Normattiva (the consolidated laws), Agenzia delle Entrate and INPS — plus pinned, per-year tables for the numbers that matter: IRPEF brackets, INPS rates, forfettario thresholds and coefficienti by ATECO, and the ATECO 2025 catalogue.
It retrieves, then reasons
For each question, a hybrid search (semantic + keyword) pulls the most relevant passages. Specialised agents read them in parallel and a synthesiser composes the answer from what was actually retrieved — not from the model's general knowledge.
It cites every sentence
Each factual sentence carries an inline citation to the exact passage it came from. Open any source to read the original. No uncited numbers, no invented articles of law.
It checks its own citations
Before you see an answer, a verifier confirms that every cited source actually appears in the retrieved material. Anything that isn't grounded is flagged rather than presented as fact.
The numbers are deterministic
Tax and contribution estimates come from a transparent, rule-based engine — the same logic for everyone — not from the language model guessing. Estimates are clearly labelled as estimates.
Your data stays in the EU
Answers are generated by an EU-hosted model (Azure OpenAI, Italy North), and your content is never used to train models. Your data — retrieval, storage and the diagnostic traces we keep to debug and improve quality — stays on EU-resident infrastructure, end to end.
Building with AI agents? You can use these same sourced tax tools inside Claude, Cursor or any MCP client.TaxCompass MCP server →
See it on your own situation
Answer a few questions and get a personalized, fully-sourced setup plan for opening your business in Italy.
Build my setup plan