Side by side
| Phileas | Philter | Microsoft Presidio | |
|---|---|---|---|
| What it is | A library you embed in your application | A self-hosted redaction API, built on Phileas | A Python library, with Docker images for a REST service you run yourself |
| License | Apache 2.0 | Apache 2.0 | MIT |
| Languages | Native Java, Python, and .NET on one policy schema | Any language, over HTTP | Python, or any language through its REST service |
| Policy model | PhiSQL, a declarative language that compiles to a versioned JSON policy you can review | The same policies, managed through Philter’s API | Recognizers and operators configured in code or YAML, with no compiled policy artifact |
| Detection | Patterns, dictionaries, and PhEye models trained to find names | The same, with a name model loaded by default | Patterns plus general-purpose spaCy, Stanza, or transformer NER |
| Conditional redaction | Built in. Redact based on zip code population, age thresholds, confidence, or patterns | Same as Phileas | Left to your own code |
| Redaction strategies | Mask, redact, hash, AES-GCM encryption, FF3 format-preserving encryption, date shift, consistent pseudonymization, synthetic replacement | Same as Phileas, plus audited re-identification of encrypted values | Replace, redact, mask, hash, AES encryption, and custom operators. No format-preserving encryption or date shift |
| Built in, rasterized so no text is recoverable under a redaction | Same as Phileas | Through the image redactor | |
| Image, OCR, DICOM | Not a focus | Not a focus | Yes, presidio-image-redactor |
| Structured data | Free text focus | Free text focus | Yes, presidio-structured |
| Deployment | Inside your own process | AWS, Azure, and Google Cloud marketplaces, a container, on premises, or air-gapped | You build and operate the service |
| Cost | Free | Free to self-host, or from $0.49 an hour on a cloud marketplace | Free |
| Support | Community, with commercial support from Philterd | Community, with commercial support and consulting from Philterd | Community only |
| Surrounding toolkit | Philter Scope (precision and recall), Philter Workspace (review), Phinder (discovery), Phield (monitoring), Philter AI Proxy and Philter MCP (LLM traffic) | Standalone | |
Service vs library
Where Phileas and Philter differ from Presidio
Presidio and Phileas are both good detectors, and Presidio is ahead on images and structured data. The difference shows up in the policy model and in what happens to a value once it is found. That is where teams moving a Presidio prototype into a governed production pipeline tend to hit walls.
A policy you can review
Policies are versioned JSON, written directly or in PhiSQL, and kept in version control. Presidio is configured in code or YAML.
Conditional redaction
Redact only zip codes below a population threshold, or only ages over a cutoff. In Presidio, that logic is yours to write.
Consistent and reversible replacement
The same person gets the same stand-in throughout a document. Values can be encrypted, including format-preserving, or date-shifted.
Java, Python, and .NET
Phileas runs natively on all three with one policy schema. Presidio is Python, or a service you call.
Models trained for PII
Name detection runs on PhEye models trained for PII. Presidio uses general-purpose NER models.
A toolkit around it
Philter Scope scores a policy’s precision and recall, and Philter Workspace adds human review. Presidio stands alone.
When you want a service instead of a library
Many teams start with Presidio in a notebook, then wrap it in their own API so other systems can call it. At that point they are building and running a redaction service, with its scaling, monitoring, and upgrades.
Philter is that service, already built. It is Phileas behind a REST API with API keys, audit logging, and policy management. You can launch it from the AWS, Azure, or Google Cloud marketplace and pay for it on your cloud bill, which is often faster than a new vendor contract. It also runs as a container on premises or air-gapped. The code stays Apache 2.0, and commercial support and consulting are available from the people who wrote it.
Moving from Presidio
The migrate from Presidio guide maps Presidio recognizers to Phileas filters and anonymizer operators to Phileas strategies and PhiSQL, with Presidio code next to its Phileas equivalent.
Where Presidio is strong
Presidio’s strengths are real, and in a few areas it is ahead of Phileas and Philter today.
- A large community and a familiar name. It is Microsoft-backed and MIT-licensed. You run
pip install presidio-analyzerand have a wide set of recognizers out of the box. - Image, OCR, and DICOM de-identification. The
presidio-image-redactorpackage redacts PII burned into images and supports DICOM medical imaging. Phileas and Philter work on text and PDF. - Structured and semi-structured data.
presidio-structuredtargets tabular and JSON data. Phileas and Philter are built around free text. - Easy Python extensibility. A custom recognizer is a small Python class, and there is a clear path to plug in transformer or LLM recognizers alongside spaCy or Stanza.
If your problem is image or DICOM de-identification, tabular data, or a Python prototype you want running this afternoon, Presidio is a good choice. All three tools share the property that matters most. They are open source, you can read the code that touches your data, and none of them sends that data to a vendor.
We want this comparison to be accurate and fair. Open source projects move fast, so this reflects publicly documented behavior at the time of writing and may have changed since. Check the current Presidio documentation before deciding, and if you spot anything inaccurate, please let us know and we will correct it.