Scoped by workspace
Core product data is shaped around account and workspace boundaries, with application and database controls protecting the main user tables.
Privacy Architecture
Built for confidential AI work.
Designed to show where memory goes.
See how yarnnn keeps durable memory scoped, attributable, revocable, and honest about the places AI context is processed. Read the policy
OAuth
Authorized access
Scoped
Workspace boundary
Provenance
Attributed changes
Not our training data
We do not use your workspace data to train yarnnn-owned models. When you ask AI to work, relevant context may be sent to model providers under their API terms.
Revocable access
Connected assistants require authorization, carry attribution, and can be revoked. Deletion and retention controls are being expanded where durable history still has limits.
How it works
Private by structure, not just by policy.
yarnnn is built around durable memory, so privacy has to be mechanical: what is stored, who can read it, who wrote it, when it leaves yarnnn, and what can be revoked.
Memory stays workspace-shaped
Files, tasks, saved memories, and revision history are organized around your workspace instead of a loose global memory pool.
Assistant access is explicit
ChatGPT, Claude, or another MCP-capable assistant can connect only through an OAuth grant you approve and can revoke.
Changes are attributable
Saved items record whether they came from you, yarnnn, a teammate, or a connected assistant, so durable memory has a visible source.
AI data flow is named
When you ask AI to use workspace context, relevant context may be sent to model providers so the work can be done. The privacy policy states that flow plainly.
What holds today
The trust story is concrete.
This is the part that is strong enough to say plainly: scoped product data, credential encryption where credentials are stored, low-PII telemetry defaults, and visible provenance.
- Most connector credentials are encrypted at rest where stored as credentials.
- Crash telemetry is configured without default PII collection.
- The purge model has explicit levels for work history, workspace state, integrations, and account deletion.
- Revision history is durable by design so users can inspect what changed and who changed it.
Current hardening
The roadmap is part of the promise.
Durable AI memory creates real privacy tradeoffs. We name the work still being tightened, especially around private content bodies, deletion completeness, credential rotation, and retention.
- Tightening private file-body reads so content is reachable only through workspace-scoped authorization paths.
- Adding garbage collection for unreferenced private content bodies after workspace resets and account deletion.
- Expanding row-level security coverage for user-scoped tables that still depend on application-layer filters.
- Normalizing remaining API-key-like connector metadata into encrypted credential storage with rotation support.
- Adding clearer retention controls for operational telemetry, historical revisions, and AI-context minimization.