New: Philter One, a redaction box for your office, is coming soon. Learn more →

Contact Us

Tell us about your stack and the privacy problems you're trying to solve. We typically respond within one business day.

Prefer email? support@philterd.ai

Please do not enter PII or PHI in this form. If you need to share an example, use a sanitized one.

Differential privacy for PII analytics

Philter Diffuse

Philter Diffuse applies differential privacy to PII counts and aggregations, so you can answer questions like "how many SSNs flowed through this pipeline last month" with a formal bound on what the answer reveals about any one document. Early access, version 0.1.0.

Why differential privacy matters for PII counts

Exact PII counts can reveal whether a particular document was in the data, especially when a count is small. Philter Diffuse adds calibrated noise so the aggregate stays useful while what it reveals about any one document stays bounded.

AggregateRawPrivate
  • Passport numbersa small count can point to one document 3 5
  • Bank routing numbersa small count can point to one document 8 11
  • Email addresses 12,480 12,477
  • Phone numbers 9,315 9,318

Large aggregates barely move, so they stay useful; the small, risky counts get the protection. Query the private counts, not the raw ones.

Why differential privacy

Provable, not promised

Formal ε-budgets backed by mathematical proof. "Differentially private" means something specific here: not a marketing word, but a measurable property of the algorithm.

A bound per document

Adding or removing one document changes the output distribution by a bounded amount, which limits what a result reveals about whether that document was in the data. The bound is per document: someone who appears in many documents is protected less, in proportion.

Aggregate queries only

Counts of PII types, never raw records. The output preserves the population-level signal you need for analytics while bounding how much it reveals about any single document.

Tunable budget

Lower ε means more noise and stronger privacy; higher ε means less noise and stronger utility. Pick the trade-off that fits your regulatory posture, per query.

Drop-in for Philter telemetry

Point Philter Diffuse at Philter and it reads the PII counts Philter records through Philter's API, with only an API key. Measure pipeline behavior without publishing exact counts.

Open source

Every line of the noise calibration, sensitivity analysis, and budget accounting is inspectable. The math is auditable, the implementation is auditable, the privacy bound is reproducible.

Frequently asked questions

If something here isn’t covered, get in touch and we’ll answer.

What is Philter Diffuse?
Philter Diffuse applies differential privacy to PII counts and aggregations. It lets you answer questions like "how many SSNs flowed through this pipeline last month" with a formal bound on what the answer reveals about any one document. It is early-access software at version 0.1.0, so its interface may change before 1.0.
What is differential privacy, briefly?
Differential privacy is a formal definition: adding or removing one document from the input changes the output distribution by at most a bounded amount, set by a privacy budget (epsilon). Philter Diffuse adds noise calibrated to that bound, which limits what someone holding the result can infer about whether a specific document was in the data.
What kinds of questions can I ask it?
Counts of PII types, such as how many documents contained an SSN, never raw records. The output preserves the population-level signal you need for analytics while bounding how much it reveals about any single document.
How do I choose the privacy strength?
The trade-off is set by the noise scale (epsilon). Lower epsilon means more noise, stronger privacy, and less utility; higher epsilon means less noise, weaker privacy, and more utility. You pick the balance that fits your regulatory posture, and you can pick it per query.
Does it protect people or documents?
Documents. Philter's counts add at most one per PII type for each redaction, so the bound covers one document. Someone who appears in many documents can move a count by as many documents as they appear in, and their protection weakens in proportion. Validate your settings against how your own data is structured.
How does Philter Diffuse fit with the rest of the toolkit?
It reads the PII counts Philter records, through Philter's API, or a JSON map of entity type to raw count. It privatizes those counts so you can share statistics about your own redaction telemetry without publishing exact counts.
Is Philter Diffuse open source?
Yes. The noise calibration, sensitivity analysis, and budget accounting are all inspectable, open source under the Apache License, version 2, on GitHub. The math is auditable, the implementation is auditable, and the privacy bound is reproducible.

Ready to use Philter Diffuse?

Grab the open source and run it yourself, or work with our team directly. Pick the path that fits.

See your options