Scoped access
Role-based workspace sessions, scoped provider keys, audit events, and revocation controls.
Control plane
Data sovereignty, made operational
Effective: July 23, 2026
Deployment boundaries, retention choices, export paths, and deletion. Not a badge, a set of controls.
Control follows the data
Farka separates workspaces and associates training, billing, and audit evidence with that workspace. Data residency follows the regions configured for application storage, object storage, GPU providers, logs, backups, and deployment targets. See processors and hosting for what each category of provider receives and where it sits.
Explicit retention
Choose whether training data and model artifacts are retained; retained storage stays visible and metered.
Deletion paths
Delete generated datasets immediately. Submit tracked user or workspace deletion from Settings → Data control.
Residency per run
Pin a run to the EU or the US and GPU selection is filtered before cost, with no fallback outside the region. Farka does not imply a region it cannot verify.
From ingest to erasure
- Ingest: selected files enter configured workspace storage.
- Train: the chosen provider receives only material needed for the job. A run pinned to the EU or the US fails rather than relocating when in-region capacity is unavailable.
- Deliver: exports follow the configured local or object-storage destination.
- Retain or delete: retention is selected per quote; dataset deletion and tracked rights requests remain available.
Enterprise sovereignty
Need a specific country, cloud account, private network, customer-managed keys, or dedicated GPU fleet? Contact us. Before production data arrives, the deployment plan should name each storage location, processor, transfer safeguard, backup window, and deletion owner.