STEP 01
WHERE IT STAYS
Train locally
Each approved party trains on its own data, inside its own boundary. No raw records cross to anyone else.
→APP DATA · FEDERATED LEARNING · SHARE THE UPDATES, NOT THE DATA
Train across rooms. Move no records.
Share the model updates. Not the data. Train across sites and parties without records leaving the room they belong in — and see exactly what crosses the boundary on every round.
WHERE IT RUNS
On your boundary
Each site trains in its own environment. Records never leave it.
WHAT CROSSES
Updates only
Model weight updates travel. The data behind them stays put.
WHAT YOU KEEP
Proof on the round
Who joined, what was aggregated, privacy budget left — on one record.
HOW IT WORKS
Three steps. No raw transfer.
A competitor lost four terabytes — including who its workers were. Keep the records in the room. Train locally, share updates only, and aggregate the gain with the record included.
STEP 02
WHAT TRAVELS
Share updates only
Model updates leave the site. The data does not. Secure aggregation and a privacy budget keep the spend-down visible.
→STEP 03
WHAT YOU KEEP
Aggregate the gain
The round closes with a review record: who joined, what was aggregated, participant health, and the privacy budget left.
ONE ROUND
Approvals in. Updates back. Proof out.
A round runs like any other release: approve the parties, aggregate the updates, gate the result, and keep the privacy budget with the record.
WHAT COMES OUT
What every round leaves with.
The data never moves. The proof does. Every round leaves a record the next one has to clear.
01
Model updates
The weight updates each approved party submits — never the records behind them.
↳ ARTIFACT
02
Privacy budget
Differential privacy spend-down and remaining headroom, written down on every round.
↳ ARTIFACT
03
Aggregation log
Who joined, who dropped, what crossed the boundary, and the secure aggregation status.
↳ ARTIFACT
04
Signed round
An identity-verified record of the round and its verdict, attached to the release.
↳ ARTIFACT
WHERE IT FITS
Five stages. Federation owns Remember.
Test the run. Review the hard cases. Recruit the right specialist. Remember what each party can share. Approve what leaves review. Federation is the Remember stage — it learns across rooms while every record stays where it belongs.
01
Test
02
Review
03
Recruit
04
Remember
● THIS PAGE
05
Approve
RELATED MODULES
More from App Data.
TRAINING
Train on what your team trusts.
Fine-tune with the rubric, the reviewers, and the data you already keep.
See the page →RL ENVIRONMENTS
Practice in a room that remembers.
Deterministic environments for agents that need to be tested before they ship.
See the page →COMPLIANCE MONITORING
The record builds as you work.
Access, audit, and retention written down without crossing the wall.
See the page →FEDERATED LEARNING
Train across rooms. Move no records.
Bring the parties. Bring the boundaries. We handle the rounds, the privacy budget, and the proof.