Boring by design. Isolated by architecture.
Workspace isolation is enforced in the database, not in application code. We're building toward SOC 2 - and we label what hasn't shipped yet, honestly.
RLS workspace isolation
Row-level security scopes every query to its workspace. Cross-workspace access is structurally impossible, not just policy.
TLS everywhere
Encrypted in transit across every surface - workspace, API, and SSE streams.
No training on your data
Your documents ground answers; they are never used to train models. Full stop.
API key scoping
Workspace-scoped API keys; sessions isolated per app; revoke and reissue any key in seconds.
Human-in-the-loop
Plan-mode proposals wait for sign-off; agent tool use is allowlisted per agent.
Run logs & audit trail
Every session, run, and approved plan is timestamped and reviewable in the console.
Inference region pinning
Pin a supported region. Google can stay in Canada. Claude's Canada pin runs in the US.
Drafts wait. Humans decide.
Customer-facing actions pause at an approval gate — a person clears them before anything runs. Every run keeps reviewable history in the console.
Agent drafts the reply
refund request · ticket #4,821
Held at the gate
nothing sends itself
Approved by Priya N.
sent, and logged
hitl.approve @priya
audit.append run #9,304
Ask us the hard questions.
Security review, data-processing terms, or residency in your own cloud - we'll answer straight.
Or email contact@sundaypyjamas.com - a founder replies.