# Likvid > Continuous close for European SMBs. Likvid reads a company's bank movements, its issued and > received invoices from the tax authority's own record, and its invoicing tool, and matches every > movement to the document behind it. What it cannot explain becomes one question for a person — > and that answer is kept, with its scope, and reused. Likvid runs on Hungarian rails today: real-time invoice reporting to the tax authority (live since 2018) and PSD2 bank data through a licensed account information service provider. The same two rails are being switched on across Central and Eastern Europe on a published schedule, which is why the product is built here first. What it is not: it is not an accountant. Classification, the VAT return and the official close stay with a professional. It is not a payment initiator — it never moves money; an approved payment run is a list a person takes to their own bank. Likvid is also a server for the AI tools a user already has: it speaks the open Model Context Protocol (MCP) at https://app.likvid.ai/api/mcp with OAuth sign-in and a per-company grant. A connected tool reads live figures with their sources and can ask for a proposal (a project closure, an advance invoice, a proforma, a reversal); the proposal waits as a card in Likvid, and there is no way to approve it from the tool. No connected tool can move money. The product is in beta and free during the beta. Access is by request. ## Product - [Home](https://likvid.ai/): what the product is, in short — the four steps, the four kinds of knowledge behind an explained movement, and links into every page below. English. - [A day in it](https://likvid.ai/day): what the product does overnight and what the five minutes in the morning look like — the morning picture, one queue with one question each, and the work prepared for approval. - [How it learns](https://likvid.ai/how-it-learns): the four kinds of knowledge (read, told, inherited, guessed), what an answer carries (scope, evidence, authority), what it remembers (lessons a person approved, precedent shown next to a similar case, preferences the user asked it to keep — never the conversation itself), and an explicit list of what the product cannot learn. - [The economic event](https://likvid.ai/economic-event): the unit everything is organised into — invoice, payment, fee, reversal and certificate as legs of one happening, which is how missing documents become countable. - [The books](https://likvid.ai/books): the double-entry ledger — trial balance, journal, ageing — what closing a month means here, the month's closing package (cover sheet, trial balance, journal, balance sheet and P&L, task list) and the tax authority's own document list compared with the books. In beta. - [Sources](https://likvid.ai/sources): what gets connected and what each source gives — the tax authority's invoice record, tax current account and the company's standing there, the taxpayer register (partner tax-number validity, VIES), the VAT-return draft where the company switches it on, bank over PSD2, invoicing tools, card processors, mailbox — including named limits (SimplePay and Barion publish no payout listing; bank consent expires roughly every 90 days). - [Your own AI](https://likvid.ai/your-ai): what a connected AI tool (Claude, Claude Code, Cursor) can read, what it can propose, how access is granted and revoked, and what it cannot do by construction. - [Continuous close](https://likvid.ai/continuous-close): the category — what changes when the explaining moves to the day the money moves, and where our numbers stop. ## Company - [Who it's for](https://likvid.ai/who-its-for): four shapes of company, the five moments that start the conversation, and an honest list of who this is not for. - [Why now](https://likvid.ai/why-now): the 2018 Hungarian invoice rail, the 2027 VAT filing change, and the country-by-country e-invoicing mandate schedule across Europe. - [Trust](https://likvid.ai/trust): the six rules enforced in code, the split between what the agent does and what the deterministic engine does, evidence instead of confidence scores, data handling, and an explicit list of what the product does not do yet. ## Magyarul - [Magyar főoldal](https://likvid.ai/hu): ugyanaz a termék magyarul, arra a piacra, ahol ma fut. - [Kinek szól](https://likvid.ai/hu/kinek): négy cégalak és az öt pillanat, ami elindítja. - [Miért most](https://likvid.ai/hu/miert-most): a 2018-as számlasín, a 2027-es áfabevallás-váltás és a régiós menetrend. ## Legal (Hungarian) - [Privacy notice](https://likvid.ai/adatkezeles) - [Terms](https://likvid.ai/aszf) - [Sub-processors](https://likvid.ai/alvallalkozok) ## Contact - Gaál Stratégia, Budapest — ngaal@ngconsult.io