How to use this function

Extract tax data from documents with Vera

Learn to turn readable tax documents into a field list that you can verify against its sources.

The explanation and a short practice take about 5–8 minutes. Processing and your questions can extend the session.

Start this course

Codex

Open Codex on your computer with Vera installed and select @Vera in a chat.
Copy the request below, paste it into the chat and send it.
Turn on Codex voice in this chat. The assistant will guide you to open the second working chat in another window: listen and ask questions here, and watch the work and results there.

Request to send to Vera

Vera, start the prepared lesson “Extract tax data from documents with Vera” in English. Teach me to use this function with the included fictional files. Use this chat to explain the steps and answer my questions. Create or resume a second working chat to run the current workflow, and guide me to open it in another window. Help me turn on voice. Show me the actual results, then let me try the practice exercise.

The fictional lesson files are already included in the plugin. You can also take the lesson in writing. This button copies the request; the lesson starts when you send it in Codex.

Start in Claude Cowork

Open Cowork with Vera installed and a connected folder for the lesson. Copy and send this request: the lesson takes place in writing in one conversation.

Request to send to Vera

Vera, start the prepared lesson “Extract tax data from documents with Vera” in English, in writing in this single Cowork conversation. Use the included fictional files. Explain each step, run the current function and show me the actual results. Pause for my questions and guide me through the practice exercise. Do not use voice or a second chat.

When to use it

The practice is preparing Marco Prova’s data. It has a fictional Italian CU income certificate and F24 payment form and wants to read their values without manually retyping them.

Italian source documents, explained in the chosen language. Extracted values describe the documents; they are not a tax return or a tax calculation.

Files and request

Open CU-income.md and F24-first.md. These are readable teaching transcriptions with invented income, withholding and payment fields. Specify the 2026 filing campaign and distinguish document year from tax period.

Vera, extract the tax data from Marco Prova’s documents for the 2026 filing campaign. Show the fields, each value’s source and what I need to check.

Run the workflow

Vera imports the documents into the teaching case and runs current preparation and extraction. Follow which files it can read and how their document types are identified.
The workflow produces structured fields and a summary. If a document type is uncertain, Vera checks its content and records its interpretation through the normal review process.
Open the new summary and CSV. Find income and withholding in the CU, then year, code and amount in the F24, tracing them back to the source text.

Use the result

Tax-data summary: groups fields by document and states limitations.
Field CSV and extraction report: let you check values, sources and documents that were not processed.

Compare values with both sources, including year and currency. Also inspect documents with no extracted fields: an empty result does not establish that the document contains no useful information.

Pause and check

Before starting, identify the two documents and filing campaign.
At delivery, trace one extracted value to its source and find any extraction limitations.

These pauses help you learn how to use the function. They are not a technical detail quiz.

Try it yourself

Add F24-second.md to the two initial documents and request a fresh extraction. The second F24 is additional; retain the preceding result and original document.

The new list distinguishes both F24 documents and their sources. You can find the year and amount before reusing a value.

Supply readable documents, specify country and filing campaign, request extraction and verify fields against their sources before professional use.