Why I am building NoAutonomo.ai
I paid €59 plus VAT for Xolo Global, around €72 a month. For my bookkeeping needs, that felt expensive. I decided to find out which parts I could handle myself and started building NoAutonomo.ai.
I want to understand where every figure in my books comes from: which document supports it, what was taken from that document and why a particular decision was made. That is why I am building a system for storing documents, reviewing entries and preparing data for tax calculations.
From an invoice to a tax return
Bookkeeping involves recurring work: collecting documents, recording income and expenses, checking them and calculating totals. Some of the work of filling in a tax return can be automated. For example, Spain’s tax agency, AEAT, offers Pre130: it reads accounting books in the required format and fills in the corresponding sections of Modelo 130. AEAT’s description of the service.
Take a fictional case: an autónomo is required to file Modelo 130, the form for advance payments of IRPF (personal income tax). There is one invoice, DEMO-001, for January to March. After review, the books show €1000 in income that counts towards IRPF. In this example, that is all the income from the business for the period.
Those €1000 go into box 01 of the first-quarter return. This box contains the relevant income from the start of the year to the end of the reporting quarter. The rule appears in AEAT’s instructions.
That entry is easy to trace: invoice DEMO-001, a line in the books and an amount in a box on the return. For the next quarter, income must be counted from the start of the year. This is one step in completing the form; calculating the final payment requires the other information in the form, including expenses and withholdings.
The example shows which work can be given to a program. The user checks that the document belongs to the right transaction and period, and the program adds up the reviewed amounts. Whether an expense is deductible or a tax rule applies needs separate consideration.
What NoAutonomo.ai does
The system already has income and expense entries, links to source documents and accounting calculations. You can open an entry and compare it with the document. Extracted information goes through review: a person confirms the facts and the accounting decision before the transaction is included in a calculation.
For the invoice in the example, this means checking the number, dates and amount before the entry affects the total. The issue date, the period when the work was done and the payment date need to be distinguished. Errors or missing information need to be resolved during review.
A reviewed and posted entry contributes to the working calculations. Submitting a return to AEAT remains a separate step. The capabilities described here refer to the project’s code; an installed version may differ.
I am developing the project around that review. I need to be able to work backwards from a total to the entries and documents, understand the basis for a decision and check it again. The same applies to the program itself: the available code lets you see how a calculation works.
Why use AI here?
In this approach, AI can help parse a document or explain a rule. The information it produces needs to be checked against the original, and a tax explanation needs to be checked against the primary source and the circumstances of the case. Several models giving the same answer does not, by itself, prove that the answer is correct.
The program calculates amounts from the accounting data. AI assistance is useful when it leaves something clear to review: extracted fields, an explanation or a link to a rule. If the answer does not provide enough grounds for a decision, the question remains open.
Who I am building it for
I am starting with Russian-speaking autónomos in Spain, especially IT professionals who want to understand their own bookkeeping and are willing to work through it.
The project’s code is available on GitHub. I want to keep the tool free for managing your own books; the terms for using the code still need to be formalised. External AI services may require separate subscriptions or API keys, and those costs need to be considered.
In the community, you can ask questions, check examples and suggest improvements. You can help by spotting an unclear step in the instructions or describing a problem using fictional data. You do not need to install the system to take part.