AI Potluck
Back to Gap Map Model components / Federated learning

Federated Compute Platform

Google
source available / Overall score: 2.0

Google's libraries for running federated programs and computations in the cross-device setting. The repository holds the client-side code and the service and data-format definitions that run in production in Google's federated learning infrastructure, and a client for confidential federated computations that verifies TEE-hosted server applications by remote attestation. The production server is not published.

At Google, the computations this code runs are written in TensorFlow Federated and compiled into deployable artifacts; TensorFlow Federated is its own record on the map.

Openness

2 high confidence
2.0
license
Apache-2.0(OSI
source
partial(client libraries and proto definitions only

What Google publishes is the device side of its cross-device platform under Apache-2.0. The production server is not open source, and the reference server that used to be in the repository has been removed, so nobody can run the platform end to end from this code.

Adoption

1 low confidence
1.0

The platform publishes no package; its client ships inside Google's apps rather than to developers. Stars are the fallback and read attention only.

Capability

3 medium confidence
3.0

The published code is the phone side of a production cross-device platform, including attestation of the servers that aggregate. Without the server half it is level with Plato.

  • https://raw.githubusercontent.com/google-parfait/federated-compute/HEAD/README.md recorded 2026-09-27

    README: "This repository hosts infrastructure for running federated computations in the cross-device setting"; a client-side implementation of the ConfidentialAggregations protocol "which uses remotely attested server-side applications that run in a Trusted Execution Environment (TEE)"; the production server is not open sourced.

Verified 2026-09-27