RECITALS Cryptography Manager

The RECITALS Cryptography Manager (CrM) is a modular Python library and REST service that brings privacy-preserving and cryptographic operations to the RECITALS platform behind a single, uniform interface. Rather than implementing cryptography itself, it wraps vetted open-source backends and adds what a platform component needs around them: parameter validation, privacy budget accounting, a sanitised audit trail, and token-based authorization.

It can be used either as a Python library, imported directly into a data processing pipeline, or as a containerised REST service that other platform components call over HTTP.

Supported Techniques

• Differential privacy: Adds calibrated statistical noise to aggregate queries so that the presence or absence of any single record has a negligible effect on the result. The strength of the guarantee is controlled by the privacy parameter ε, and every query's cost is deducted from a per-user budget that the service enforces.
• Authenticated encryption: Protects data at rest and in transit using vetted AEAD primitives, binding each ciphertext to its context so it cannot be silently replayed elsewhere.
• Key management: Generation, serialisation, loading and rotation of keysets. Rotation retains previous keys, so data encrypted before a rotation stays readable afterwards.

Implementation Status

Operation	Backend	Library	REST	Status
Differential privacy	PyDP	yes	yes	Implemented
Encryption / decryption	Google Tink	yes	yes	Implemented
Key management	Google Tink	yes	yes	Implemented
Homomorphic encryption	Pyfhel	no	no	Not implemented — returns 501
Secure multi-party computation	MPyC	no	no	Not implemented — returns 501

Homomorphic encryption and secure multi-party computation are recognised operations with a settled request shape, so adding them means writing one adapter each rather than reworking the interface.

The Cryptography Manager delegates all cryptographic work to established open-source libraries, so that the privacy and security guarantees rest on implementations that have been reviewed far more widely than this component could be.

Core Libraries

PyDP A Python wrapper around Google's differential privacy library, providing ε-differentially private aggregate statistics with noise calibration and sensitivity analysis inherited from a well-reviewed implementation.

Google Tink A misuse-resistant cryptographic library. Algorithm parameters, nonce generation and ciphertext tagging are handled internally, and only vetted primitives are reachable, leaving little room for the implementation errors that break real systems.

Python Infrastructure

The manager leverages a modern Python stack for performance and reliability:

• FastAPI: The REST service and its generated OpenAPI documentation.
• Pydantic: Request, response and configuration validation.
• PyJWT: Bearer token signature and expiry verification.
• HTTPX: Forwarding audit records to downstream platform services.
• Loguru: Structured application logging.
• Pytest: Automated testing and quality assurance.

Contributors

Name	Role	Contact
Konstantinos Chousos	Development	kchousos@di.uoa.gr
Dimitrios Pavlou	Development	dimpavlou@di.uoa.gr
George Stamoulis	Supervision	gstam@di.uoa.gr

Funding

This work is part of the RECITALS project, funded by the European Union under Horizon Europe grant agreement No. 101168490.