Flows

A flow is the definition of one document type: the fields you want extracted, the hints that tell the AI where to find them, and everything attached downstream. It is the object in Tavnit that turns a document into data — Collections, Subjects, Splitters and Cleaners all exist to feed or refine what a flow produces.

There is no template to draw and no coordinates to map. You describe the fields in plain terms and the flow works across layouts, so one Supplier invoices flow can handle twenty vendors whose invoices look nothing alike. That is also why the schema and its hints are where nearly all extraction quality is won or lost.

Click Create Flow on the Flows page and Tavnit asks “How do you want to start?”. All three paths end in the same editable flow, so pick whichever gets you to a good first draft fastest.

Tavnit templates cover invoices, purchase orders, receipts, bank statements, delivery notes, lab results, contracts, quotes, credit notes, ID documents and resumes. If you answered the welcome questions when you set up your organization, the templates that match your documents are shown first, and the empty Flows page offers to create them all in one click with Create these flows.

Every path finishes with a name (at least 3 characters) and a description (at least 10 characters). The flow opens in the builder with a short “Set up your flow” guide, and it is Active from the start — use the switch in the top bar to deactivate it. An inactive flow cannot be run.

Every field is one of two kinds, decided by whether the value appears once per document or once per line. This is the most consequential choice in the schema: it determines the shape of your webhook payload, your Bucket rows, your CSV and your review screen.

A flow with only metadata fields — an ID document, a certificate, a one-page form — returns a single row per document holding those values, so the result still lands in your table view, Bucket, webhook and email like any other run.

A flow can also stamp System Columns onto every row it outputs — the flow ID, the flow name and the run ID. Turn these on when several flows write into one Bucket and you need to know which document a row came from.

Each field carries a type. Types are not cosmetic: they decide whether a value sorts, sums, compares in a Cleaner rule and charts correctly. Setting them right at the flow saves work at every later stage.

Hints are how you disambiguate a field without writing a template. They matter most when a document contains several values that look alike — three dates, two totals, an order number and an invoice number in the same header block.

For a metadata field you can combine any of these:

For a table field the hints are different, because the AI is locating a column rather than a point on the page:

Example values are skipped for Date and Image fields, so put any date guidance in the additional-hints box instead.

Some table columns hold several values in one cell — a size-and-quantity breakdown like S:2 M:5 L:3 printed in a single box. A composite field splits that cell into sub-fields so each part becomes its own value instead of a string you have to parse later.

1. 1Add a table field and switch Composite Field on.
2. 2Define the sub-fields that make up the cell — for example size as text and quantity as a number.
3. 3Add example values showing how the grouping appears in your documents.

Sub-fields hold single values, so they can be Text, Number, Date or Mixed/Alphanumeric — but not Image. Composite fields are only available on table fields, because the whole point is unpacking a repeating cell.

An Image field pulls a figure out of the document rather than text: a product photo, a signature, a stamp, a logo. A metadata image field holds one image per document; a table image field gives each row the image belonging to it.

• Images are stored privately, so they arrive downstream as time-limited links rather than raw bytes
• The same kind of link appears in the webhook payload, the email output and the CSV cell; an export from the run page generates fresh links
• Links in deliveries are valid for seven days — download the image rather than storing the link
• Image fields cannot be composite or accept multiple values, and example-value hints do not apply to them

Build metadata fields first, then table fields, then hints. Test against a real document early — a schema that looks right on paper and a schema that survives your actual paperwork are different things.

1. 1Click Create Flow and pick a template, an AI suggestion or a blank flow. Give it a name that describes the document (Supplier invoices) and a real description — it improves extraction accuracy and is what a Collection matches on later.
2. 2Under Metadata Fields, click Add Field for each value that appears once per document, setting the data type as you go.
3. 3Under Table Fields, add a field for each column of the repeating line-item table.
4. 4Add extraction hints only to the fields that need them.
5. 5The flow is already Active. Click Run, process one real document, and compare the result against the source.

Drag fields by their handle to reorder them.

A flow accepts PDFs, images and spreadsheets, through every route — upload, email and API. Tavnit decides how to read each file, so you do not configure anything per format.

Very large sheets are rejected with a clear error. To process each sheet of a workbook as its own document, send it through a Splitter instead.

The schema decides what comes out; the rest of the flow decides what happens to it. Each item below is independent — attach only what you need, in any order, at any time.

Email Trigger options. Once the trigger is on, the panel shows the flow's inbox address and two settings. Allowed Senders limits who can start a run by email; leave it empty to accept any sender. Process Email Body also extracts from the message text itself, either Only when nothing is attached (the default, so a cover note never starts a second run) or Always. The sender's address is recorded on each run and shown on the Runs page. Email Integration covers the details.

Once a flow has processed some documents, the Diagnose button in the flow's top bar opens the Flow doctor. It studies your recent completed runs, finds the fields that keep coming back empty or with the wrong type, and proposes corrections.

1. 1Open the flow and click Diagnose. The button appears once the flow has at least one field, for Owners and Admins.
2. 2Tavnit reads the recent runs, re-reads a few of the documents where fields failed, and drafts corrections. This takes a moment.
3. 3Review the result. Each problem field shows how often it was empty or had the wrong type, with the Current hint beside the Proposed one. It may also offer a Suggested description.
4. 4Keep or edit the fixes you agree with and click Apply. Nothing changes until you apply; closing the panel discards the proposals.

If everything looks fine you will see “Extraction looks healthy”. Diagnose only runs when you click it. Re-run a real document after applying fixes to confirm they helped.

The Runs page lists every run in your organization; each flow also has its own Recent Runs tab. Use Filter to narrow by status, date range, flow, user, source or a pasted run ID.

The Triggered By column shows who started the run — the user for an upload, or the sender's address for a run that arrived by email. The Source column says how it arrived: manual upload, API, email and so on.

Most failures are the document's fault — an unreadable file or an unsupported type — and retrying would not help. A few are not: a processing worker restarts mid-run, or an AI provider has a temporary outage. For those, an organization can have Tavnit retry on its own.

1. 1Go to Settings → Organization → Failed runs. Only the Owner can see this tab.
2. 2Turn on Automatically retry failed runs.
3. 3Set Maximum retries per run to 1, 2 or 3.

• A retried run keeps the same run ID, so API polling, webhooks and links keep working.
• Outputs only fire when a run completes, so a retry never delivers a document twice.
• Never retried: invalid files, unsupported types, cancelled runs, runs paused for review, and runs created more than a day ago.
• A run lost to a worker restart is detected automatically. With auto-retry on it is queued again; with it off, it is marked Failed.
• The run page lists Previous attempts, each marked Lost worker or Failed with its error, and the Runs list shows a retry badge.

To hear about runs that end in failure, add addresses under Run Failure Notifications in the same tab.

When a field comes back wrong, the fix is nearly always in that field's definition rather than in the document. Start with Diagnose if the flow has runs, then work through these — the first two solve most cases.