01
Check supported identifiers
Rules and validators detect supported patterns, including selected Swiss formats. Coverage depends on the policy and input.
← All products
Local data minimisationDeveloper preview
Bowden-PII
Useful context. Less personal data.
Replace supported personal identifiers locally before an application sends text onward. Keep the original-value mapping under your application’s control.
The reason
A useful request does not always need a real name.
Applications often send identifying details alongside the context a model actually needs. Reducing that exposure should happen before the request leaves the system.
Bowden detects supported identifiers and replaces them with consistent placeholders. The mapping stays separate. This reduces exposure, but context can still identify someone, so evaluation remains necessary.
Inside Bowden-PII
What you can do.
Select an image to take a closer look.
02
Replace consistently
Repeated values keep the same typed placeholder within a call. The reduced text and replacement mapping are separate outputs.
03
Control the mapping
Your application decides how to store original values and who can restore them. The mapping needs its own access controls.
04
Evaluate before adoption
Test missed detections and unnecessary replacements on representative examples. The learned token-classification runtime is experimental.
Read the Bowden-PII research and evaluation findings →
Follow the Bowden-PII workflow
BOWDEN-PII / LOCAL DATA FLOW
Useful context. Fewer personal identifiers.
Follow a piece of text through the local runtime.
Swipe to explore · choose a step below ↔
The task needs context. Not every identity.
Your application passes the text to Bowden in your own environment. The local runtime does not need an external AI service for this step.
BEFORE · FICTIONAL INPUT
Please confirm Thursday’s appointment with [email protected]. Send updates to [email protected].
AFTER · REDUCED TEXT
Please confirm Thursday’s appointment with [EMAIL_1]. Send updates to [EMAIL_1].
Inspect the local mapping and audit metadata
Mapping · local only
- [EMAIL_1]
- [email protected]
Audit · no original values
{
"policy": "strict",
"total_spans": 2,
"labels": {
"EMAIL": 2
},
"sources": {
"rule": 2
},
"conflicts": [],
"raw_values_in_audit": false
}Prepared examples from the rule-based runtime · strict policy. No text is processed by this website. Replacing identifiers reduces exposure; it does not guarantee anonymity.
Availability, scope and technical resources
Bowden is a developer preview. Start with a representative sample, evaluate missed detections and unnecessary replacements, then agree the runtime and integration. Detection coverage depends on the supported categories and the selected policy.
Placeholder replacement reduces exposure; it does not guarantee anonymisation or legal compliance. Context may still identify someone. The mapping contains original values and requires its own access and storage controls.
Technical questions
Is the resulting text anonymous?
Not automatically. Bowden replaces detected identifiers with placeholders. Remaining context can still identify people. Evaluate detection, residual context and the receiving system for your use case.
Is it production-ready?
Bowden-PII is in early development. The deterministic core is available. Evaluate missed identifiers and false positives on your own data before production use.
Does this website process my text?
No. The illustration uses prepared fictional examples. Use the local Python runtime to process your own text.
What is available?
A deterministic core, three policy presets, Python API, CLI and audit metadata without raw values. Neural detection is experimental.
Who manages the original values?
Your application. Access control, secure storage, tenant separation and controlled recovery are integration responsibilities.
Bowden-PII guides