ManagedCode.Presidio.Core 0.0.1

ManagedCode Presidio for .NET

Status: Early alpha. The API surface area is evolving while we complete the port from the Python reference implementation.

ManagedCode Presidio is the .NET 9 rewrite of Microsoft's Presidio project. The original Python codebase remains the authoritative reference implementation and is vendored in this repository as a Git submodule under external/microsoft-presidio. Our goal is feature parity with the Python release while delivering first-class NuGet packages for .NET applications.

Solution Layout

• Presidio.slnx – Visual Studio / dotnet solution for all libraries and tests.
• src/ManagedCode.Presidio.Core – cross-cutting primitives (TextSpan, RecognizerResult, AnalysisExplanation, etc.).
• src/ManagedCode.Presidio.Analyzer – contracts for recognizers and NLP artefacts.
• src/ManagedCode.Presidio.Anonymizer – operators, engine results, and shared anonymization abstractions.
• src/ManagedCode.Presidio.ImageRedactor – image-domain types (bounding boxes, request/response models).
• src/ManagedCode.Presidio.Structured – helpers for structured/semi-structured payload anonymization.
• tests/ManagedCode.Presidio.* – unit and integration suites validating parity against the Python behaviour.
• external/microsoft-presidio – Git submodule pointing at the upstream Python repository.

NuGet Packages

The solution produces the following packages (versioned centrally via Directory.Build.props):

• ManagedCode.Presidio.Core
• ManagedCode.Presidio.Analyzer
• ManagedCode.Presidio.Anonymizer
• ManagedCode.Presidio.Structured
• ManagedCode.Presidio.ImageRedactor

Publishing is handled by the release.yml GitHub workflow which runs on pushes to main.

Getting Started

Prerequisites

• .NET SDK 9.0.301 (configured via global.json).
• Git 2.35+ with submodule support.
• Optionally, Python 3.9+ if you need to run the original reference code or regenerate fixtures.

Clone the Repository

git clone https://github.com/managedcode/presidio.git
cd presidio
git submodule update --init --recursive

Build & Test

dotnet restore Presidio.slnx
dotnet build Presidio.slnx --configuration Release
dotnet format
dotnet test Presidio.slnx --configuration Release

We always run dotnet format before dotnet test to ensure consistent styling across the solution.

Usage Examples

The .NET APIs mirror the Python surface. The snippets below show the most common entry points.

Anonymizing raw text

using System.Collections.Generic;
using ManagedCode.Presidio.Anonymizer;
using ManagedCode.Presidio.Core;

var recognizerResults = new[]
{
    new ManagedCode.Presidio.Core.RecognizerResult("PERSON", new TextSpan(11, 16), 0.85),
};

var anonymizer = new AnonymizerEngine();
var result = anonymizer.Anonymize(
    "My name is James Bond",
    recognizerResults,
    new Dictionary<string, OperatorConfig>
    {
        ["PERSON"] = new OperatorConfig("replace", new Dictionary<string, object?>
        {
            [ReplaceOperator.NewValueKey] = "Agent"
        })
    });

// result.Text == "My name is Agent Bond"

Working with collections (BatchAnonymizerEngine)

using System.Collections.Generic;
using ManagedCode.Presidio.Anonymizer;

var batch = new BatchAnonymizerEngine(anonymizer);

var response = batch.AnonymizeDict(new[]
{
    new DictRecognizerResult(
        "names",
        new[] { "John", "Jill" },
        new[]
        {
            new[] { new ManagedCode.Presidio.Anonymizer.RecognizerResult("PERSON", 0, 4, 0.9) },
            new[] { new ManagedCode.Presidio.Anonymizer.RecognizerResult("PERSON", 0, 4, 0.9) },
        })
});

// response["names"] == ["<PERSON>", "<PERSON>"]

Deanonymizing previously encrypted entities

using System.Collections.Generic;
using ManagedCode.Presidio.Anonymizer;

var deanonymizer = new DeanonymizeEngine();

var decrypted = deanonymizer.Deanonymize(
    cipherText: "My name is S184CMt9Drj7QaKQ21JTrpYzghnboTF9pn/neN8JME0=",
    entities: new[] { new OperatorResult(11, 55, "PERSON") },
    operators: new Dictionary<string, OperatorConfig>
    {
        ["DEFAULT"] = new OperatorConfig("decrypt", new Dictionary<string, object?>
        {
            [EncryptOperator.KeyParameter] = "WmZq4t7w!z%C&F)J",
        })
    });

// decrypted.Text == "My name is Chloë"

Python Reference Submodule

The external/microsoft-presidio directory tracks the upstream Python implementation. Tests under tests/ManagedCode.Presidio.Core.IntegrationTests load fixtures derived from the Python suite to prove behavioural parity of key primitives. As the port progresses, more fixtures will be mirrored directly from the submodule so that regressions can be caught early.

If you update the submodule revision, ensure you run:

git submodule update --remote external/microsoft-presidio

and commit the resulting SHA change.

Roadmap

1. Core parity – finalise domain primitives and validate them against Python fixtures.
2. Analyzer engine – port recognizer registry, pattern recognizers, and ONNX-backed NER execution.
3. Anonymizer engine – reproduce the operator pipeline, conflict resolution, and policy parsing.
4. Image/Structured pipelines – deliver feature parity for image redaction and structured anonymisation flows.
5. Packaging – stabilise public APIs, documentation, and automate NuGet publishing.

Progress is tracked in AGENTS.md.

Contributing

We welcome pull requests and issues. Please read CONTRIBUTING.md and follow the existing coding conventions (see .editorconfig). All code must target .NET 9 and pass dotnet format + dotnet test before a PR can be merged.

License

This project is licensed under the MIT License.

Maintainers

ManagedCode SAS is the primary maintainer of this fork. For questions, please open a GitHub issue or contact the ManagedCode team.