Getting Started

Tavnit reads documents and gives you structured data back. You describe the fields you want once, send it invoices, receipts, purchase orders or forms, and get typed rows out — without building a template per layout or writing any parsing code.

Extraction is the starting point rather than the whole product. Once the data exists, it can be normalised, checked by a person, stored, delivered to your systems, or handed to an agent that acts on it. The rest of these docs cover those stages; this page covers the first one.

Six words cover almost everything in Tavnit. Learn what each one owns and the rest of the documentation reads much faster — most confusion comes from mixing up flows, Collections and Splitters, which do three different jobs.

A flow is the schema for one document type. Name it after the document rather than the project — Supplier invoices, not Q1 automation — because that name and description are what a Collection later uses to route documents to it.

1. 1On the Flows page, create a new flow and name it.
2. 2Upload a sample document. Tavnit suggests the fields it can see, which is faster than typing them from scratch.
3. 3Add, rename or delete fields until the schema is exactly what you need.
4. 4Switch the flow to Active.
5. 5Send one document through and check the result.

Flows walks through each of these steps properly — field kinds, data types and the hints that tell the AI where to look.

Tavnit distinguishes values that appear once per document from values that repeat. That single distinction decides the shape of everything downstream — your webhook payload, your Bucket rows and your CSV all follow it.

Each field also has a type — text, number, date, mixed or image — and getting it right matters more than it looks: a total typed as text will not sum, compare or chart. Flows covers the full schema in depth, including extraction hints, composite fields and how to fix a field that comes back wrong.

Four ways in, all producing the same kind of run. Start with a manual upload to prove the flow works, then switch to whichever route matches how documents actually reach you.

Open any run to see the extracted fields beside the source document, plus the log of what happened during processing. That log is the first place to look when a result is not what you expected.

Tavnit bills in credits. Extraction is charged per page, so a ten-page PDF costs ten credits whether it produces one row or two hundred. The other operations have their own rates.

Once extraction works, the next step depends on what is wrong with the data or what you need to do with it. These are the three most common directions.