How HELM Protects Your Data
HELM OS is built on Supabase (PostgreSQL, Auth, and Edge Functions) with row-level security enforced on every table that holds customer data. This page describes, honestly and specifically, how workspace isolation, AI approval gates, and data protection work today.
Workspace Isolation
Every workspace's leads, tasks, pipeline data, conversations, and AI memory are isolated at the database layer using PostgreSQL Row Level Security (RLS) policies scoped to workspace membership — not just application-layer checks. A user in Workspace A cannot query, and Supabase itself will not return, data belonging to Workspace B, regardless of what the client application requests.
Multi-tenant separation is enforced the same way for every table added since HELM's founding, including the most recent additions (Standing Instructions and the Actions/Decisions observability log).
AI Actions Are Never Silently Autonomous on Customer-Facing Work
Approval gates by design
HELM's Decision Engine classifies every AI-proposed action by risk. Customer-facing actions — sending an email, posting to Slack, anything a third party would see — always require explicit human approval before they execute, in every execution mode (Guided, Assisted, or Autonomous). This is a hard invariant in the code, not a setting you can turn off.
Full observability
Every AI decision and action — what was proposed, why, its risk level, and its outcome — is logged to an internal audit table. Admins can review this history for their own workspace at any time.
Verified vs. inferred data
When HELM's AI surfaces information it hasn't directly verified (e.g. an inferred company detail), it is labelled as such in the interface — never presented as confirmed fact.
Standing Instructions can't remove safety
Workspace owners can set standing rules (e.g. "always ask before emailing a deal over €50k"). These rules can only add approval requirements on top of the defaults — never remove a safeguard a high-risk action already has.
Data Protection & GDPR
HELM OS is operated in compliance with GDPR and Irish data protection law. Full detail — what we collect, why, your rights, and how to exercise them — is in our Privacy Policy. In short: you can request a full export or permanent deletion of your workspace data at any time from within your account settings, and deletion requests are honoured within 30 days.
Integrations We Connect To
HELM connects to third-party services only with your explicit authorisation via OAuth, and only requests the minimum scope needed for the feature you enable. Current integrations include Gmail, Google Calendar, Slack, Stripe, HubSpot, and QuickBooks. See the full list on our Integrations page.
Email sync is thread-based and privacy-safe by design: HELM only fetches threads it already knows are related to outreach it sent, and never scans your inbox generally. Message bodies from your own drafts are not stored server-side beyond what's needed to send — subjects only are retained for reference.
Questions About Security?
If you have a specific security question, a vulnerability to report, or need a document for your own vendor review process, email us and we'll respond directly.