Policy: A named set of rules that tells Philter which entity types to find (names, SSNs, email addresses, dates, and so on) and what filter strategy to apply to each (redact, replace, mask, encrypt, hash, and others). Every filter request names the policy to apply; Philter ships a policy named default. Policies are authored as JSON; this client treats the policy body as an opaque JSON string.
Context: An arbitrary label used to group requests (for example, by tenant or job). It is passed on each filter request and is echoed back on the response.
Document ID: Philter assigns an identifier to each filtered document and returns it in the X-Document-Id response header; the client exposes it on the response object.
Consistent pseudonymization: Replacing a value with a stand-in, and using the same stand-in for that value everywhere. Philter records the mapping as entries on the context, so the property holds across documents and requests that share a context. This is what preserves referential integrity: relationships and joins in the redacted output still line up, because one source value has exactly one replacement.
Re-identification: The CRYPTO_REPLACE and FPE_ENCRYPT_REPLACE strategies are reversible, value by value, through reidentifyValues(...). reidentifyText(...) restores every CRYPTO_REPLACE value in a text, such as an LLM's response to a redacted prompt. Redaction, masking, and hashing are not reversible. See Re-identification.