Built to be trusted with
everything a family owns.

Row-Level Security - isolation in the database engine

Tenant isolation is implemented via Postgres Row-Level Security scoped to your office, so cross-tenant reads are blocked inside the database itself - not merely filtered in application code. Internal server-only tables are locked to the service role outright. This isolation has been tested through internal adversarial review (RLS coverage re-audited in July 2026, gaps closed the same day); an independent external penetration test is pending. Admin/cron paths use a service role that never reaches the browser.

Layered authentication

Password, magic-link, WebAuthn passkeys (Face ID / Touch ID / hardware keys), and TOTP two-factor, on top of Supabase Auth. Role-based access - Owner, Advisor, Accountant, Beneficiary, plus scoped team-member roles - is enforced server-side on every API route; billing changes are owner-only.

Encryption everywhere

AES-256 at rest across database, backups and document storage; TLS 1.3 in transit. Vault documents are private objects served only through short-lived signed URLs to authenticated members of your office.

Append-only audit trail

Every material action - asset changes, document access, plan changes, sign-ins - is recorded with actor, timestamp and before/after state. The audit log is append-only for application users and tenant roles, including the owner; it is enforced by application design and access controls rather than a database-level immutability guarantee.

Hardened application surface

Security headers and CSP allowlists on every connection and frame source, with inline/eval script allowances (required by our framework and Plaid Link) tracked openly as a limitation rather than presented as a strict CSP. HSTS, X-Frame-Options DENY, nosniff, restrictive Referrer and Permissions policies. All inbound webhooks (Stripe, BTCPay, meeting transcription) are HMAC-signature-verified with constant-time comparison and fail closed. Scheduled jobs require a secret bearer token.

Data integrity engineering

The pipeline refuses to record or display implausible data: garbage snapshots are rejected at capture, structural breaks quarantine stale history, and implausible feed values are held for human review instead of silently applied. Each layer between a provider and your screen has an explicit guard, and each guard is covered by tests.

EU data residency

Primary database in Frankfurt, Germany (eu-central-1). Vault documents mirrored nightly to a separate, independent project in Ireland (eu-west-1). Every copy of your data at rest stays inside the European Union; the few non-EU sub-processors are disclosed above.

No custody, by principle

AEGISfo never holds client funds, private keys, or trading permissions. Bank feeds are read-only tokens, wallet tracking uses view-only public keys, card payments live inside Stripe, and Bitcoin settles to self-custody. The design goal is that a compromise of AEGISfo could expose information - never move money.

Managed PostgreSQL holding all structured data, encrypted at rest with AES-256. Row-Level Security enforces tenant isolation inside the database engine itself. Supabase Auth handles sign-in (passwords, magic links, WebAuthn passkeys, TOTP 2FA); the Family Vault lives in Supabase Storage under private, signed-URL access.

Runs the Next.js application and every API route as isolated serverless functions over TLS 1.3, plus the scheduled jobs (sync, snapshots, FX, billing, backups). Secrets live exclusively in Vercel environment variables - never in the codebase.

Card payments happen on Stripe-hosted checkout pages. Card numbers never touch AEGISfo servers - we hold only an invoice reference and a signed webhook confirmation. Webhook signatures are verified with constant-time HMAC comparison before any invoice is marked paid.

Bank connections use Plaid Link: your credentials are entered with your bank or Plaid - never on, nor visible to, AEGISfo. We store only a scoped access token that can read balances, and revoke it at the provider when you disconnect.

Read-only brokerage connections. Positions and balances flow one way - into your portfolio. No trading permissions are ever requested.

Powers Aegis Insights and document extraction via API. Under Anthropic’s commercial API terms, your data is not used to train models. Prompts carry the minimum context needed (portfolio summaries, document metadata) - never credentials or keys.

Delivers invoices, statements and alerts. Emails contain summaries and links back into the authenticated app - sensitive detail stays behind sign-in.

On-chain balances are read from public blockchain data using addresses and extended public keys you provide. Public keys can only VIEW - they cannot move funds. AEGISfo never holds, requests, or stores private keys. Not your keys, not your coins - so we keep no keys at all.

Bitcoin invoices will be handled by our own BTCPay Server connected watch-only to our treasury via an extended public key: payments land directly in self-custody. No payment processor ever holds the funds.

A note on AI and your data

Aegis Insights runs on Anthropic’s Claude models via their commercial API, under terms where your data is not used to train models. Prompts contain the minimum needed - portfolio summaries, document metadata, extracted text - never credentials, keys, or payment details. AI output is always labelled and framed as research rather than advice. We apply prompt-injection hardening and ongoing adversarial testing (data-not-instructions boundaries, grounding rules, regression tests) - prompt injection cannot be fully solved, so we treat it as a continuous practice, not a finished feature.

We claim

• The controls on this page are implemented in code and configuration, and are re-audited as the platform evolves.
• Our infrastructure providers hold the provider-level certifications listed in the registry above - those are their attestations, covering the layers they operate, and they do not constitute AEGISfo certification.
• Authorization boundaries, tenant isolation, webhook verification and data-integrity guards are covered by automated tests that run on every change.
• Adversarial internal reviews (backend, frontend, AI prompt-injection) have been run, and every finding was either fixed or is tracked openly.

We don’t claim - yet

• AEGISfo itself has not yet completed its own SOC 2 audit - provider certifications do not transfer, and we won’t imply otherwise. A first-party attestation is planned as the platform matures.
• No independent third-party penetration test has been commissioned yet. Internal adversarial reviews are not a substitute, and we don’t present them as one.
• No system is “100% secure”, and we will never use that phrase. Security here is a practice - layered controls, honest accounting of gaps, and rapid fixes - not a badge.