Replace¶

Replace mode replaces each detected entity with an alternative token. Anonymizer provides four strategies.

Strategy comparison¶

Strategy	Example output for `"Alice"` (first_name)	Deterministic	Generated by LLM
Substitute	`Maya`	No	Yes
Redact	`[REDACTED_FIRST_NAME]`	Yes	No
Annotate	`<Alice, first_name>`	Yes	No
Hash	`<HASH_FIRST_NAME_3bc51062973c>`	Yes	No

Substitute¶

Replaces entities with LLM-generated synthetic values that are contextually plausible. This is the only replacement strategy that requires an LLM call.

from anonymizer import AnonymizerConfig, Substitute

# Default: uses the replacement_generator model
AnonymizerConfig(replace=Substitute())

# With additional instructions
AnonymizerConfig(replace=Substitute(
    instructions="Replacement IDs must always start with the same 4 characters as the original."
))

Field	Default	Description
`instructions`	`None`	Additional instructions for the LLM replacement generator.

Model role

Substitute uses the replacement_generator role. Override it if required in your model config:

selected_models:
  replace:
    replacement_generator: your-model-alias

Redact¶

Replaces entities with a label-based marker. The original text is removed entirely.

from anonymizer import AnonymizerConfig, Redact

# Default: [REDACTED_FIRST_NAME]
AnonymizerConfig(replace=Redact())

# Custom template
AnonymizerConfig(replace=Redact(format_template="****"))

# Template with label only
AnonymizerConfig(replace=Redact(format_template="[{label}]"))

Field	Default	Description
`format_template`	`[REDACTED_{label}]`	Template with optional `{label}` placeholder.
`normalize_label`	`True`	Uppercase and clean the label before substitution.

Annotate¶

Tags entities with their label but preserves the original text. Useful for review and debugging.

from anonymizer import AnonymizerConfig, Annotate

# Default: <Alice, first_name>
AnonymizerConfig(replace=Annotate())

# Custom template
AnonymizerConfig(replace=Annotate(format_template="[{label}: {text}]"))

Field	Default	Description
`format_template`	`<{text}, {label}>`	Template with `{text}` and `{label}` placeholders. Both are required.

Note

The original text is still present, so this is not privacy-safe on its own.

Hash¶

Replaces entities with a deterministic hash digest. The same entity text always produces the same hash, enabling consistent de-identification across records.

from anonymizer import AnonymizerConfig, Hash

# Default: <HASH_FIRST_NAME_3bc51062973c>
AnonymizerConfig(replace=Hash())

# Short digest with SHA-1
AnonymizerConfig(replace=Hash(algorithm="sha1", digest_length=8))

Field	Default	Description
`algorithm`	`sha256`	Hash algorithm (`sha256`, `sha1`, or `md5`).
`digest_length`	`12`	Number of hex characters to keep (6--64).
`format_template`	`<HASH_{label}_{digest}>`	Template with `{digest}` required; `{label}` optional.