AI Potluck
Back to Gap Map Model components / Federated learning

P2PFL

P2PFL
open core / Overall score: 2.0

P2PFL is a Python library for decentralized federated learning, in which peers exchange model updates over a peer-to-peer network with gossip protocols instead of reporting to a central server. The same code runs as an in-memory or Ray simulation or across machines over gRPC, where mutual TLS admits only nodes holding certificates from a shared authority. It trains PyTorch, TensorFlow/Keras and JAX models and can add local differential privacy to outgoing updates.

Openness

4 high confidence
4.0
license
GPL-3.0(OSI
source
public(the library, its gRPC and in-memory protocols and its examples)
core features withheld
yes — the vendor's plan table leaves "Advanced features" out of the free Community (Open Source) plan and includes them in an Enterprise plan it marks as launching soon

P2PFL's library is GPL-3.0, but its maker's own plan table holds back "advanced features" from the free open-source plan for a paid Enterprise plan, which is open core. That plan is still marked as launching soon, and the page does not say which features it withholds.

Adoption

1 medium confidence
1.0

PyPI downloads of `p2pfl`, the library's own package.

Capability

3 medium confidence
3.0

P2PFL runs a real federation between machines, with mutual TLS so that only nodes holding certificates from the network's authority can join, and local differential privacy on each update. Operators distribute the certificates and start each node themselves, since the command-line node launcher is still marked as coming soon, which keeps it a step below Flower.

Verified 2026-09-28