Continuous audit. Zero critical findings. Multi-tenant by construction.

Multi-tenant RLS from day one

Every table that holds tenant data has Row Level Security policies on read AND write. Cross-tenant access is impossible at the database layer, not just the app layer.

BYOK with envelope encryption

Customer-provided model API keys are envelope-encrypted (AES-256-GCM under a per-row data key, itself wrapped under a platform master key). Plaintext keys never leave the gateway process and are never logged.

Security gates in CI on every push

Every push and PR runs typecheck, lint, unit tests, a cross-tenant isolation test, a production build, gitleaks secret-scanning, and a competitor/upstream-leak gate. Any failure blocks merge. A deeper LLM-assisted SAST sweep across 312+ files (DeepSec) runs as a periodic audit; reports are persisted and auditable.

No competitor-name leakage CI gate

A pre-commit script scans every customer-facing surface for banned competitor names. Founder doctrine: beat them silently — names cannot leak into copy, code, commits, or comments.

Prompt-injection ward on every external read

Every agent that ingests externally-sourced content (web pages, third-party repos, MCP tool outputs, RAG retrievals) wraps that content in untrusted-source tags before the LLM sees it. Suspicious instructions are surfaced to the operator, not executed.

Sandboxed agent execution

Risky agent runs (anything that touches the shell, the customer's files, or external services) go through a sandbox runner rather than the host process.

POPIA consent gate enforced in code

Outbound customer email is gated by a consent + suppression check before send (assertSendable), not just policy. Backed by a POPIA monitor that audits consent capture, purpose limitation, minimisation, retention, lawful basis, and cross-border transfer.

SOC 2 Type II

ISO 27001

POPIA Information Officer registration

VAPID / Push security