Skip to main content
Ingestly charges credits at two moments: once when a document arrives, and again for the nodes a workflow run executes on it. This page lists both.

Document ingestion

Every document that Ingestly accepts is charged 1 credit per page as soon as it has been parsed, whether it came through an Upload, Email or Webhook trigger or landed in an inbox. The charge does not depend on what happens next: a document that runs immediately, one that waits under Held and one that waits Unassigned in an inbox all pay the same on arrival. That is the trigger node’s price: an Upload, Email or Webhook trigger costs 1 credit per page, collected when the document arrives rather than when a run starts. The internal Forward trigger is free, because its documents were already paid for when they first arrived. If your organization cannot cover the document’s pages, the document is not ingested: it appears as Failed, nothing is charged, no run starts, and a notification tells you why. Add credits and upload or send it again. Credits a running workflow has reserved do not count as available, so a document can be refused while the balance still shows enough; it clears once that run settles. An organization without an active subscription is refused the same way, and the notification says so instead of asking for credits. Deleting, unassigning or expiring a document does not return the ingestion credit. It paid for receiving and parsing the file, which already happened.
Each email attachment is its own document, so a single email with two 2-page attachments charges 2 attachments × 2 pages × 1 credit = 4 credits on arrival. A single 5-page upload charges 5 × 1 = 5 credits.

Actions

Actions process documents and data within the workflow.

Parse

Chaining parse and extract on the same document runs OCR twice and bills both. Skip the parse node when extract is the next step; extract already handles OCR internally. Only chain them when you need the parsed text for a separate purpose (routing, archiving, multiple schemas off one OCR pass). See parse best practices.

Extract

Engine 2 has no precision setting; it’s a single-tier engine.

Classify

By example includes the AI comparison of each page with its examples’ sample documents; it costs no more when AI is unavailable and word similarity decides instead.

Filter

Split

The 0.5-credit tier kicks in only when the node has to read text content to make its decision (keyword matching, a page description, or auto boundary detection). Rules that test only on page numbers stay free.

Validation

Other actions

Outputs

Output nodes deliver results. All outputs are free.

Schema inference

Using the Generate button inside an extract node to infer a schema from a sample document also consumes credits. The charge is applied per page processed during generation.

Assistant chat

The in-app assistant is billed by usage rather than by page: each reply’s token usage converts to credits. A short question costs a small fraction of a credit; a long session working through several tool steps costs more, on the order of ten credits. Fractional usage accumulates per organization and appears in the transaction history in whole credits, as Assistant chat entries. Usage that has not appeared in the transaction history yet still counts against your available balance, so it moves as you chat rather than only when it settles.
When available credits reach zero, the assistant stops answering until credits are added, either by upgrading or at your next renewal.
The assistant on the marketing site is free and does not draw on your credits.

Worked examples

These examples show how to add up credits for realistic workflows. Numbers assume single-page documents unless noted; multiply by the page count for longer documents.
Office and CSV documents are read as text rather than rendered into pages, so they always count as a single page however long they are. A 400-page Word document and a one-line CSV cost the same.

Minimal extraction workflow

A typical “submit a document, get structured data back” flow:
A 5-page invoice costs 15 credits: 5 when it arrives, 10 when the run settles.

Branching workflow with classify, review, and validation

A workflow that classifies documents, extracts on a focused schema, validates the result, and pauses uncertain runs for human review:
For an invoice (matches the first branch): For a non-invoice (matches the other branch): The other branch never executes extract, so non-invoices cost a fraction of the matched ones. Both branches paid the same trigger credit when the document arrived.
The other branch costs 1.2 credits per page and the invoice branch 3.2, the first credit of each paid on arrival. You are billed for the branch that actually ran, not for the more expensive one it might have taken.
To estimate a monthly bill, multiply the per-page total for each branch by the average pages per document and the volume you expect to see in that branch.

How a run is billed

A run is billed for the steps that actually executed, and its cost is the sum of those steps. Nothing else is added, and nothing is charged for a node that never ran. When a run starts, Ingestly reserves what it could cost at worst: it walks the workflow, takes the most expensive path the branches allow, and holds that many credits. A hold is not a charge. It comes out of what you can spend on the next run, so two runs cannot both spend the same credits, but it never leaves your balance. Your Subscription page shows held credits separately from your balance for exactly this reason. When the run finishes, the hold is released and you are charged for what ran. If the cheap branch ran, you pay for the cheap branch. A few consequences worth knowing:
  • The trigger is paid on arrival, and the first run shows it. The 1 credit per page leaves your balance when the document is parsed, before any run. The first run on that document reports it on its trigger step and in its total, so the run reads what the document really cost; a later run of the same document shows 0 for the trigger because nothing more was charged. A run that fails at its very first step adds nothing to that.
  • A run that fails or is cancelled is billed for the steps it completed before it stopped. Those steps did real work and cost real money to run.
  • A test run is billed the same way. Testing a single node bills that node.
  • A loop bills its body once per iteration. A loop running an extract over 50 line items costs 50 extracts. The loop node itself is free; what is inside it is not.
  • Each sub-workflow call is billed on its own. A called workflow is a run in its own right, and it appears as its own row on the Usage page.
  • Repeating a document you have already processed still costs credits. Ingestly caches extraction results, which makes the second run faster, but the cache saves machine time rather than money and the step is billed at its normal rate.
Because a run’s bill is the sum of its steps, the Usage page can show you exactly which node in which workflow spent your credits, down to the individual step.

Credit allocation

New organizations start with 200 welcome credits so you can build and test a workflow before subscribing. That grant is one time: the Free plan is not a monthly allowance, so the 200 credits are not topped up each month. On a paid plan, credits are allocated monthly with your subscription and unused credits do not roll over. See subscription for plan details and credit balances.

Per-workflow usage

Each workflow page shows a daily credits chart for the last 7 days, so you can spot which workflows are consuming the most credits and which days had spikes. Every step records what it was charged, and a run’s bill is the sum of those charges, so the run detail page breaks the cost down node by node and the numbers add up to the total.

Subscription

Plan tiers, credit allotments, and balance

Alert rules

Get notified when your credit balance drops below a threshold

Parse best practices

Avoid double-OCR and pick the cheapest engine that works

Extract best practices

Match engine and precision to your documents