Understanding the Token System

What is a token?

A token is a usage credit. We deduct a small number of units each time you scan for PII, swap in placeholders, or restore the originals. You always know the cost up front. You only pay for operations that succeed.

Heads-up: on anonym.legal the word “token” means two different things. This page is about billing credits — the credits you spend on operations. Elsewhere on the site we also talk about tokenized PII, which is the technical placeholder (like <PERSON_1>) that replaces a name in your text. Those are not billing units, and you don’t pay for them directly.

At a glance

• Free
  200/cycle
  Enough to try the platform on a few small documents.
• Basic — €3 / month
  1,000/cycle
  Comfortable for occasional document anonymization.
• Pro — €15 / month
  4,000/cycle
  Regular processing across many documents.
• Business — €29 / month
  10,000/cycle
  Heavy daily use, batch jobs, and API integrations.

Unit allocations reset at the end of each billing cycle (30 days for Free, 31 days for paid plans). Unused credits do not roll over.

What spends tokens?

Free actions: checking your token balance and getting a cost estimate cost zero credits. You can preview the price before any paid call.

Real costs by document size

Measured on English text with dense PII (the entity count is what the analyzer actually found). These are the numbers your account is charged in production:

Rule of thumb: For short inputs the analyze charge floors at ~4 tokens (driven by the base cost plus ~0.1 token per enabled entity type). Above 1 000 characters at GUI defaults the cost scales at roughly 1 token per 312 characters.

These numbers are auto-derived from the production test suite — they update whenever the test JSONL reports are regenerated and pass replay.

Does the operator matter?

Replace, redact, hash and mask all cost the same. Encrypt costs more because every encrypted entity adds the 0.8 × encryptCount term to the apply step — about 3× higher at the 160-character input below, shrinking to ~1.2× at 50 000-character documents (where the per-character term dominates). All rows below use the exact same 160-character input + 14 selected entity types.

Replace, redact, hash and mask all cost the same — pick whichever fits your privacy goal. Encrypt costs ~3× more at this input size because it adds the 0.8 × encryptCount term.

Does language matter?

Characters and detected entities drive cost, not language. The table below shows the same content — a name, email, and phone number — across all 48 supported languages. Every variant lands at the same token cost when chars and entity filter match. Character counts still vary widely between scripts, since CJK languages encode the same information in fewer characters.

Running low? Top-ups

On paid plans you can buy extra credits mid-cycle. Free plan users don’t have top-ups — upgrade to Basic to enable them.

Top-up credits expire at the next cycle reset. They don’t roll over either.

Quick questions

Do unused units roll over?

No. Allocations and top-ups both reset at the end of each cycle.

What happens if I run out mid-task?

The next call returns a 402 with the exact shortfall. Buy a top-up or wait for the cycle to reset.

Do failed operations cost credits?

No. Tokens are only deducted after an operation succeeds.

Does the choice of operator change the price?

Replace, redact, mask, and hash all cost the same. Encrypt is the only exception — it adds 0.8 × encryptCount to the apply step, making it ~3× more expensive at small inputs (and ~1.5× at large documents where the per-character term dominates).

Why are some operations cheaper than others?

Analyze and anonymize do detection work (the expensive step). Decrypt only walks the saved mapping, so it’s much cheaper.