Security, plainly stated
Your customer conversations are the most sensitive thing you’d hand us. Here is exactly how they’re protected.
inrelay stores every workspace in Supabase-hosted PostgreSQL with row-level security on every table, encrypts data in transit with TLS and at rest by default, and sends nothing to AI providers that they can train on. Attachments live in access-controlled storage served through signed URLs.
Built on a database that enforces the rules
inrelay runs on Supabase-hosted PostgreSQL, and the database - not the app code - is the authority on who can see what. Every table carries row-level security policies scoped to your workspace, so a query from one tenant physically cannot return another tenant’s rows. Every change that matters - assignments, live-chat lifecycle, message delivery - is recorded once and audited.
The same discipline applies to what customers can see. Your team sees the full internal chronology of a live chat; the customer-facing widget receives only neutral lifecycle facts. Assignee names, handoff notes, internal workflow changes and private notes never cross that boundary.
The practices.
Four things we hold ourselves to on every feature we ship.
Per-workspace tenancy
Row-level security scopes every read and write to your workspace. Threads, contacts, knowledge base, widget settings - all of it is partitioned at the database layer.
In transit and at rest
All traffic between the apps, the widget and the backend runs over TLS. Data is encrypted at rest by default on Supabase’s managed infrastructure. Attachments are served through signed, expiring URLs from storage buckets with their own access policies.
Least-privilege by design
Teammates only reach the workspaces they belong to, and privileged operations run through security-definer database functions that check membership first. The desktop app opens external links in your own browser, never inside the app.
AI that doesn’t train on you
Drafting, triage and search indexing use commercial model APIs whose terms state that content sent through the API is not used to train their models. The AI sees only the conversation and your knowledge base, and nothing it writes sends without a human’s approval unless you turn Autopilot on - and even then refunds, upset customers and VIPs always stop and wait for a person. The optional live-chat chatbot answers visitors directly and hands off to your team when it cannot help.
Your data is yours
That’s not a slogan, it’s a policy with two teeth. You can export your conversations, contacts and knowledge-base articles whenever you want them, and when you delete your account we delete your data - not “archive” it, not keep it for a rainy day. We don’t sell it, share it with advertisers, or mine it for anything beyond running your inbox.
See it before you trust it.
Twenty minutes, real conversations, every question welcome.
Certifications, honestly
inrelay does not currently hold SOC 2, ISO 27001 or similar certifications, and we won’t imply otherwise with a wall of grey badges. Formal audits are on the roadmap as the product and team grow; until then, this page describes what is actually built, and we’d rather you judge us on that.
05 · DisclosureFound something?
If you believe you’ve found a vulnerability, we want to hear about it before anyone else does. Write to security@inrelay.ai with the details and we’ll respond as fast as a small team honestly can. Please give us reasonable time to fix an issue before disclosing it publicly.
Questions about anything above - data handling, the approval model, or how plans affect limits - are welcome via contact.