Security at OzBrain

Updated

OzBrain holds knowledge your agents read and write. Every claim on this page is either something you can check in your account, or an architecture explanation you can verify in the open design.

  • OzBrain does not train on your brain

    OzBrain never uses your content to train foundation models and never sells it. Connected agents process it on your subscription when they call the connector. What those platforms do with content they receive is governed by their terms on your account.

    See connected clients

  • An append-only log you can export

    Writes, promotions, disputes, and task events land in an account log you can open and export as CSV (up to 5,000 events per download). Event rows are append-only. Displayed titles follow the current article or task name when those change later.

    Open your logs

  • Per-client access you can revoke

    Agents connect through OAuth 2.1. Each client appears in Connections with revoke. There are no long-lived API keys to leak. Revoke cuts that client.

    Manage connections

  • Per-account encryption at rest

    Article and version bodies sealed under a key unique to your account. Titles and the routing index stay readable so agents can navigate. Bodies decrypt only on the request path: when your connected agent reads or writes, and during disclosed maintenance.

  • Maintenance loop

    Scheduled coherence passes that keep the brain true as agents write: decay, audit, and reconsolidation on a cadence you can see in the log.

How it works

  • Tenant isolation in the database

    Every knowledge row is account-scoped. Row-level security is forced in Postgres for bearer reads. Agent writes go through SECURITY DEFINER database functions that re-check tenant scope before any mutation. Hostile cross-tenant cases live in the automated test suite.

  • Immutable version history

    Canon writes land as version snapshots. You can inspect what changed and which actor wrote it. Staged writes promote only with an explicit decision; conflicts surface instead of silently overwriting.

  • The inference boundary

    Models see your content when your own connected agent reads or writes it through the connector. OzBrain does not mine, sell, or train on customer brain content. Your connected platform's handling follows that platform's terms.

  • Transport

    The product is served over HTTPS. Auth and data store on Supabase; the app runs on Vercel.

Who touches your data

Every party that can see content or account data in the course of providing the service. The last row is the one most tables omit: the AI platform you connect is a recipient on your account, under that platform's terms.

Sub-processors and data recipients for OzBrain
PartyRegionPurposeDataRetention
SupabaseConfigured project regionAuthentication, Postgres, and OAuth grantsAccount identity, brain content and versions, Chronicle, OAuth connections, rate-limit bucket keys (visitor IP and email identifiers for abuse caps), and connector onboarding sessions (email plus auth token hashes for visitors who have not finished signup)Account data for the life of the account. Rate-limit buckets and unfinished onboarding sessions can outlive a signup attempt; onboarding tokens stop working after expiry
VercelEdge and function regions serving ozbrain.coApplication hosting and request deliveryHTTPS traffic, standard request logs, deployment metadataPer Vercel platform defaults
NitroSendProvider regions for email deliveryTransactional and lifecycle emailEmail address, activation attributes, authentication emails with one-time action links (sign-in, recovery, email change), and About contact-form submissions (name, email, message body)While your account receives product email, or until you ask us to stop. Auth action links expire per Supabase auth settings. Contact submissions are retained for support handling
Your AI platformWherever that provider runs your clientRuns the agent you connect (Claude, ChatGPT, Cursor, and others)Whatever that client reads or writes through the OzBrain connector on your accountUnder that platform's terms, on your subscription, controlled by you

Report a concern

Security reports: security@monsef.com. OzBrain is a product of Monsef Holdings Pty Ltd, Melbourne, Australia.

One brain. Every agent.