Security & Trust

Security built into the architecture — not bolted on.

NevTech Catalyst handles the documents, policies, and AI keys that run your business. So isolation and encryption aren’t a settings page — they’re structural. Your data is separated at the database layer, your secrets are encrypted with managed keys, and every sensitive action is recorded. Here’s exactly how.

01 · Our approach

Four controls do the heavy lifting.

Most breaches come from the boring fundamentals being optional. We made them mandatory and moved them down the stack, where they can’t be skipped by a forgotten check in application code.

Tenant isolation at the database

Every workspace’s data is separated by Postgres row-level security — enforced by the database engine, not just application code.

Secrets encrypted with managed keys

API keys and connection tokens are sealed with AES-256-GCM under a Google Cloud KMS key. Plaintext is never written to logs.

Role-based access control

Owner, admin, and member roles gate every sensitive action. Admin-only operations are checked in middleware on every request.

Everything important is audited

Key rotations, document changes, connector edits, and more are written to an append-only, per-workspace audit trail with actor and IP.

02 · Multi-tenant isolation

Cross-tenant data leakage is structurally impossible.

Many SaaS platforms keep tenants apart with a WHERE org_id = ? clause in application code. One missing clause, anywhere, and data leaks across customers. We don’t rely on that. Isolation is enforced by Postgres itself, through row-level security policies the database applies to every single query.

Org-scoped tables run under Postgres FORCE ROW LEVEL SECURITY — there is no permissive fallback.
Every query executes inside a transaction that pins the current workspace ID; the policy compares each row against it.
Default-deny: a query that arrives without a workspace context returns zero rows, by design — not an error you can forget to handle.
Documents, chat sessions, retrieval chunks, stored keys, and audit events are all covered by this policy.

This same isolation extends to your document AI — retrieval can only ever surface chunks from your own workspace. See how the RAG engine respects tenancy →

03 · Encryption of secrets

Your keys are sealed with envelope encryption.

The most sensitive things you store with us are credentials — your own AI provider keys and the tokens for any tools you connect. We protect them with the same envelope-encryption pattern cloud providers use for their own systems.

Per-secret data keys

Each stored secret gets its own random 256-bit data key and is encrypted with AES-256-GCM, which also detects any tampering with the ciphertext.

Wrapped by Google Cloud KMS

That data key is itself encrypted (“wrapped”) by a key held in Google Cloud KMS that never leaves the KMS boundary. Decryption requires a privileged, audited KMS call.

Never logged, zeroed after use

Plaintext secrets are never written to logs, and the in-memory data key is zeroed immediately after each operation.

In transit, everything runs over HTTPS. At rest, the underlying database and file storage are encrypted by Google Cloud Platform.

04 · Authentication & access

Verified identity, least-privilege access.

Identity. Sign-in is handled by Google Firebase Authentication, which issues short-lived, signed identity tokens verified on every API call. The platform is stateless — there are no long-lived server sessions to steal.
Roles. Workspace roles (Owner / Admin / User / Viewer) determine what each person can see and do. Sensitive actions — storing keys, editing connectors, changing roles — require Admin or Owner.
Least privilege. Public, shareable assistants run on a restricted path that cannot reach your internal admin documents or governance policy — only the material you explicitly attach to them.
05 · Network & connections

Connections can’t be turned against your network.

Catalyst lets you connect external tools and data sources. Any feature that fetches a URL on your behalf is a potential server-side request forgery (SSRF) risk — so we guard it directly.

Private addresses blocked

Connected servers are validated before use: internal/private IP ranges and localhost are rejected, and the hostname is re-resolved to defeat DNS rebinding tricks.

HTTPS enforced

In production, connected endpoints must use HTTPS — no plaintext callbacks, no downgrade.

Rate limiting & hardened headers

The API applies per-client rate limiting and standard hardened security headers on every response.

06 · Audit trail

A record of who did what, in your workspace.

Sensitive actions across the platform — storing or rotating a key, changing a member’s role, editing a connector, lifecycle events on documents and skills — are written to an append-only audit log scoped to your workspace. Each entry captures the actor and originating IP, and the log is protected by the same row-level isolation as the rest of your data.

Append-only — entries are recorded, never quietly edited.
Per-workspace — you only ever see your own activity.
Resilient — auditing never blocks the action it records, and failures are surfaced, not swallowed.
07 · Your data & AI

Your business data isn’t training anyone’s model.

Not used for training

Your business data is never used to train AI models. It’s used to answer your questions, in your workspace, and nothing more.

Bring your own keys

You can run AI calls on your own provider keys (Anthropic, OpenAI, Google). They’re stored with the envelope encryption described above and used only for your workspace’s requests.

Embeddings stay yours

Document embeddings used for retrieval are isolated per workspace under the same database-level policy — they’re never pooled across customers.

08 · Compliance & ownership

Certified where it counts — and you own your data.

SOC 2 Type II

NevTech Catalyst is SOC 2 Type II certified — independently audited controls for security, availability, and confidentiality.

HIPAA

HIPAA-compliant with a Business Associate Agreement available on the Scale plan, for teams handling protected health information.

You own your workspace

Everything in your workspace is yours. On cancellation, we delete it permanently within six months.

Building your AI governance posture is a first-class part of the platform, too — the AI Conformance Statement gives you a documented, versioned policy to hand to insurers and auditors. See the governance layer →

Security you can put in front of IT

Bring the platform to your security team.

Tenant isolation at the database, KMS-encrypted secrets, audited activity, and SOC 2 Type II + HIPAA. Start with a free assessment, or talk through the details with us.