Security & data

Your data stays yours: isolated per organization, accessed by role, and accounted for change by change.

Tenant isolation

Every workspace is a separate tenant. Files, embeddings, ingested tables, chats, agents, and audit logs are all scoped to your organization, and that scope is enforced on every request. A query physically cannot cross into another tenant's data.

Authentication & access

• JWT-based sessions with short-lived access tokens and refresh rotation.
• Roles. Admins manage members, billing, and connectors; members use the product. Superuser support actions (account emulation) are explicit and stamped into the audit log as "emulated by …".
• Connector credentials are stored per connector, used only for syncing, and never exposed back through the API.
• Databases you connect should use a read-only user; on top of that, generated SQL is read-only, injection-checked, and row-capped on our side.

Audit trail

Every mutating request in the platform is recorded automatically:

• Who: the acting user (and the emulating superuser, if any).
• What: method, path, status, and a bounded copy of the request payload.
• Redaction first. Sensitive fields (passwords, tokens, secrets, credentials, API keys) are replaced with [redacted] before anything is stored. File-upload bodies are never buffered or stored, only noted.

Agents, flows, and dashboards additionally expose their own scoped audit views, so you can see the change history of a specific thing next to its run history.

Your data and models

• No training on your data. Workspace content is used only to answer your team's questions and run your agents.
• Grounding over generation. Answers are composed from retrieved passages and executed queries; numbers come from SQL results, not model prediction.
• Deletion. Removing a file or connector removes its content from your knowledge base; connector re-syncs prune items deleted at the source.

Questions?

Security review, data-processing questions, or a deployment conversation: hello@vectropic.ai.