Skip to main content

Document Processing Consumption & Pricing

When you upload documents for your agents to use, ToothFairyAI may need to run CPU-intensive document processing on them — extracting images from pages and PDFs, or running Advanced Vectorisation for higher embedding accuracy. This page explains how that processing is metered and billed in UoI.

Both features are optional and only run when you enable them for a document.


What is measured

Document processing runs on a dedicated worker (the same class of worker used for other compute-heavy jobs). Two separate features each consume compute:

FeatureWhen it runsWhat it does
Image ExtractionEnabled per document (incl. PDFs & web pages)Extracts the images present in the document/PDF and prepares them for your agents to retrieve
Advanced VectorisationEnabled per documentTable-accurate structure analysis and richer embedding of the document content

Each runs only while enabled, and each is billed only while that processing is actually running.


How it is priced

Billing is per-minute — each processing step is billed a minimum of 1 minute of compute, rounded up to the next full minute. Every charge is drawn from your UoI credit balance (the same credits you use for LLM calls, images, virtual browser sessions, etc.).

ComponentMeterMinimumCurrency
Image ExtractionWall-clock minutes of image extraction1 minute per document-processing step1 UoI = $0.01 USD
Advanced VectorisationWall-clock minutes of advanced vectorisation1 minute per document-processing step1 UoI = $0.01 USD

Indicative rate

ItemApprox. cost
Image Extraction step~0.6 UoI per minute of compute (min ~0.6 UoI)
Advanced Vectorisation step~0.6 UoI per minute of compute (min ~0.6 UoI)

A typical small document finishes each enabled step quickly, but every enabled step is billed a minimum of 1 minute (~0.6 UoI). Larger documents — many pages, or many embedded images — accrue more minutes at the same per-minute rate. The exact per-minute rate is calibrated to our infrastructure cost and may drift slightly with cloud pricing.

What counts toward the clock

  • Image extraction: rendering/reading pages, locating and cropping embedded images, and preparing them for agent retrieval.
  • Advanced Vectorisation: Docling table-accurate structure analysis and content embedding preparation.

The clock for each step stops when that step finishes.


A worked example

A PDF with several pages and many embedded images where the workspace enables both Image Extraction and Advanced Vectorisation, and each step takes ~40 seconds of compute:

Line itemAmountCost
Image Extraction (40s → min 1 min @ ~0.6 UoI/min)1 min~0.6 UoI
Advanced Vectorisation (40s → min 1 min @ ~0.6 UoI/min)1 min~0.6 UoI
Total~1.2 UoI ≈ $0.012

Running the same document with neither feature enabled charges none of the above — only the normal LLM/embedding usage applies.


Observability

Both the Image Extraction and Advanced Vectorisation compute appear as cost lines in your workspace dashboard usage view, so you can see per-document how much compute was billed and for which feature.


Summary

  • Image Extraction and Advanced Vectorisation are optional, per-document features that run CPU-intensive processing.
  • Each is billed per minute of compute, minimum 1 minute (~0.6 UoI/min), drawn from your UoI balance.
  • A document only incurs the compute charge for the feature(s) you enable on it.
  • There are no standing or subscription charges — you pay only for the document-processing compute you actually use.