Use case
Grounding an accounting AI assistant in Polish law
A use case: an accounting product answers its users' tax questions from the official text, with citations and the right version for the date, through one API.
Other languages:DeutschEspañolFrançaisPolski
An accounting product that adds an AI assistant needs the assistant to answer from the law, not from memory: the right provision, the wording in force on the date that matters, and a citation the accountant can check. VeriActa is that source layer. The product sends the question and a date to one API and gets back the provisions with their citations and links, then shows them or grounds its own assistant in them; the accountant keeps the final word.
What does the product ask?
The questions its users already type into the assistant, in their own words:
- “Do we charge VAT on this invoice to a company in Germany?”
- “What was the VAT exemption limit in 2025?”
- “Which PKD code fits a bookkeeping office?”
- “Is this NIP an active VAT payer?”
None of them names an act or an article. Finding the provision is the source layer’s job.
What does it get back?
One call to the search endpoint with the question and a date:
POST /v1/law/PL/search
Authorization: Bearer vak_…
{ "query": "VAT exemption limit for small businesses", "as_of": "2025-12-31" }
The response carries the provision quoted in full from the consolidated text, its citation (art. 113 ust. 1), a link to the official document, the version in force on as_of, any amendment already published with the day it applies, and a coverage field that says what the answer rests on. The full contract is described in OpenAPI 3.1 at /v1/openapi.json, and the example on the home page shows it with two dates.
Beyond statutes, the same key reaches the codes and registers accounting work runs on: PKD and PKWiU classifications, ZUS guides and Ministry of Finance handbooks, and the company registers by NIP or KRS.
Why not let the model answer?
Because a language model on its own gets exactly the details accountants care about wrong: it mixes up thresholds from neighbouring articles, quotes a wording that changed, or invents an article number that looks right (why that happens). Grounding the assistant in returned provisions turns each statement into something the product can show with its source, and something the accountant can verify in one click.
Is it used in production?
Yes. Through partner AI agents, VeriActa already serves more than 1,700 accounting teams. Every change to the search is checked against real questions from accountants first, in everyday and formal wording, and ships only when every correct answer stays correct.
What does it cost a product?
Credits per answered call: 2 for a provision, 10 for a search, 100 for a written answer with citations. Every account starts with 200 free credits. Monthly plans start at €12 for 10,000 credits; pay as you go costs from €2.00 per 1,000 credits, less with volume or prepaid credits; and partner plans serve products that embed VeriActa. See pricing.
Frequently asked questions
Can our own assistant call it as a tool?
Yes. The REST API and the MCP server share one contract, so an agent can call VeriActa as a tool over MCP, and your backend can call the same lookups over REST.
Can we show the cited text to our users?
Yes, on every plan, with source attribution; partner plans add redistribution.
Which countries can the product cover?
Poland, Spain, Germany, Denmark, the Netherlands, Ireland, Italy, France and Finland, Sweden, Czechia, Luxembourg, Estonia, Slovakia, Lithuania, the United Kingdom and EU law from EUR-Lex are live; the coverage by country lists all 27 member states, the United Kingdom and EU law. The API is the same for every country.