API Reference¶

Top-level package¶

pyfe4ai: Python Functional Encryption for AI Security and Privacy.

The package exports its entry points:

create() builds the key generator of a scheme given by its id "family/variant" and returns it with the scheme’s crypto class;
list_schemes(), get_scheme() (a Scheme: classes, kind, capabilities, review status), CAPABILITIES and get_scheme_status() query the scheme registry (pyfe4ai.core.registry);
backend_info() (pyfe4ai.backend.backend_info()) reports the backend in use (native extension or Python), the extension’s build, the threads, the Gaussian sampler and the parameter cache directory;
the exceptions and warnings of the library.

The scheme classes themselves are exported by the family packages pyfe4ai.schemes.sife, mife, mcfe and quadratic, the ML adapter by pyfe4ai.ml.

pyfe4ai exports the entry points of the library:

create() – the key generator of a scheme given by its id "family/variant", built from a config and not yet set up, with the scheme’s crypto class (CreatedScheme);
the queries of the scheme registry (pyfe4ai.core.registry): list_schemes(), get_scheme() (a Scheme: classes, kind, capabilities, review status), CAPABILITIES, get_scheme_status() and SchemeStatus;
backend_info() (backend_info()) – the backend the calling thread uses (the native extension or the Python implementation) and how it was chosen, the extension’s version and build, the threads per native call, the Gaussian sampler and the parameter cache directory; importing pyfe4ai does not load the extension, calling it does. Without the extension the library also says so once, with an FEWarning at the first fallback to Python (Native Backend);
the exceptions: FEError and its subclasses FEConfigError, FEKeyError, FEValidationError (also a ValueError), FESchemeError (also a RuntimeError) and MixedSharingsError (an FESchemeError);
the warnings FESecurityWarning and FEWarning;
__version__.

The scheme classes are exported by the family packages (pyfe4ai.schemes.sife, mife, mcfe, quadratic), named after their scheme’s id (see pyfe4ai.core.registry). The status and security-profile modules are documented on Security & Status.

Public – what applications use; changes are listed in the changelog (from version 0.3.0: Upgrading from 0.3.0):

pyfe4ai (above);
the family packages pyfe4ai.schemes.sife, mife, mcfe and quadratic: the key generators and crypto classes and their documented methods (the scheme modules, e.g. pyfe4ai.schemes.mcfe.ring_lwe, document them);
pyfe4ai.core.registry, pyfe4ai.core.config (the config classes and their options), pyfe4ai.core.status, pyfe4ai.core.profiles, pyfe4ai.core.exceptions and pyfe4ai.core.ids (the rules for server and client ids, with is_id and as_id);
pyfe4ai.ml and pyfe4ai.ml.quantization;
from pyfe4ai.backend: available, get_backend, set_backend, backend, active_backend, get_threads, set_threads, implementation, describe, backend_info and MISSING_EXTENSION_NOTICE;
the few entry points in otherwise internal modules that the documentation names: get_gaussian_sampler, set_gaussian_sampler and gaussian_sampler (pyfe4ai.primitives.sampling), pairing_available and DEFAULT_PAIRING_GROUP (pyfe4ai.primitives.pairing.pairing_backend), PartialDecryptionBudget (pyfe4ai.schemes.threshold.kl) and default_cache_dir (pyfe4ai.core.base).

Internal – documented here for contributors and reviewers; they may change in any version without notice:

pyfe4ai.core.base (the base classes and the parameter cache), pyfe4ai.core.formats and pyfe4ai.core.versions (the classes that describe the serialized objects and their version markers; the serialized formats themselves are documented and versioned there), pyfe4ai.core.one_time, pyfe4ai.core.forksafe (the fork handler; what threads, copies and fork do is on Threads, Processes and Copies), pyfe4ai.core.constants;
pyfe4ai.schemes.common, pyfe4ai.schemes.threshold and the helpers of the scheme modules other than their classes;
pyfe4ai.primitives and the rest of pyfe4ai.backend: the interface the scheme and primitive layers call (pyfe4ai.backend.ops(), pyfe4ai.backend.ring(), pyfe4ai.backend.interface) and its two implementations (pyfe4ai.backend.python_impl, pyfe4ai.backend.native_impl);
pyfe4ai._validation and the native extension pyfe4ai._rust_native.

The package is organised in layers:

pyfe4ai.core – base classes and the parameter cache, the configurations (pyfe4ai.core.config), the serialized objects (pyfe4ai.core.formats) and their versions (pyfe4ai.core.versions), security profiles, review status, one-time precomputations, the fork handler, exceptions;
pyfe4ai.schemes – the schemes by family (sife, mife, mcfe, quadratic), with the components several families share in pyfe4ai.schemes.common and pyfe4ai.schemes.threshold;
pyfe4ai.primitives – lattices, Gaussian sampling, hashing, groups and pairings;
pyfe4ai.backend – the operations the schemes and primitives take from a backend (pyfe4ai.backend.interface), their native (Rust) and Python implementations, the selection between them, and threads;
pyfe4ai.ml – the ML adapter (encrypted inference, federated aggregation) and quantization.