Security & Trust — Aventide

How we protect your business data — in plain language.

Aventide helps small businesses run on AI. That means you trust us with real
information about your company, your clients, and your money. We take that
seriously. This page explains, without the jargon, how your data is protected
and how we think about security.

The short version

  • Your data is encrypted in transit and at rest.

  • Every business’s data is walled off from every other business, enforced on our servers.

  • We deliberately don’t hold your most dangerous data — no passwords, no card numbers.

  • Our AI works only inside your business and can’t reach across to anyone else’s.

  • We build on infrastructure that is independently SOC 2 audited (Vercel, Supabase, Anthropic, Stripe, Cloudflare, Outseta).

  • We’ll always tell you honestly what we do and don’t have.

Your data is encrypted, coming and going

Everything that moves between you and Aventide travels over encrypted connections
(TLS/HTTPS) — someone snooping the network sees nothing usable. Everything stored
in our database is encrypted at rest with AES-256, managed by our infrastructure
providers. If a hard drive physically walked out of a data center, it would be
unreadable.

For your most sensitive items — like the access tokens that connect your Google
or Stripe account — we add a second lock: we encrypt them a second time at the
application layer, with a key kept in a separate secrets vault, not in the
database. So even someone with full database access couldn’t use them.

One business, walled off from every other

Aventide is multi-tenant — many businesses use the same platform — so keeping them
separate is the most important thing we do.

When you log in, our servers resolve exactly one thing: which business you’re
allowed to see. From that point on, every action — every piece of data read or
written, every AI request — is automatically stamped and filtered by your business
ID, enforced in our server code. There is no “show me everything” mode. A request
for another business’s data simply returns nothing.

This isn’t left to chance or to the AI’s judgment — it’s built into the data layer
itself.

How our AI actually works — it can’t wander

People hear “AI agent” and picture something with the keys to everything. Ours is
the opposite.

Every time the AI does something for you, it stays locked to your business the
entire time. It never gets a master key to “all customers.” It literally cannot
ask for another business’s data, because the only question it’s ever allowed to
ask is “…for this business.”

We also treat the information we feed the AI as inert data, not instructions.
So even if someone tried to hide sneaky commands inside their own content to trick
the AI, it’s built to read that content as information, not as orders to follow.

And our AI provider (Anthropic) does not train its models on your data under
the business terms we operate on.

We deliberately don’t hold your most dangerous data

The safest data is the data we never store:

  • No passwords. Login runs through a dedicated, SOC 2-audited provider
    (Outseta). We never see or store your password.

  • No card numbers. Payments run through Stripe (PCI-DSS Level 1 certified).
    Card data lives with Stripe; we only ever hold a meaningless reference token.

For your money area specifically: we connect to your own Stripe account (Stripe
Connect), so funds flow directly to you — never through us. In our system, a
“payment” is just a reference number, an amount, and a date that Stripe reports
back. Your invoices and expenses are your business records (amounts, categories,
dates) — never card or bank credentials.

In a worst-case breach, the blast radius is limited by design: encrypted at
rest, no passwords, no card data, and connected-account credentials
double-encrypted with a key stored separately.

You can delete, and it stays deleted

When you remove something the AI remembered about your business, we don’t just hide
it — we record a suppression marker so the AI won’t silently re-learn and re-add
it later. Deletion is honored, not cosmetic.

Built on audited infrastructure

We don’t reinvent the foundations of security — we build on providers who are
independently, continuously audited:

Provider Role Independent compliance
Vercel Application hosting SOC 2 Type 2
Supabase Database, storage, realtime SOC 2 Type 2
Anthropic AI model provider SOC 2 Type 2 · no training on your data
Outseta Login & billing SOC 2 · handles all password security
Stripe Payments PCI-DSS Level 1


How AI-written code ships here (this is a feature, not a risk)

We build with AI assistance — and we treat that as a reason for a stricter
process, not a looser one:

  • Every change is reviewed by an independent set of eyes before it ships — the
    one who writes a change and the one who reviews it are never the same. AI-written
    code doesn’t get a shortcut; it gets an extra gate.

  • Every change must pass an automated test suite of thousands of checks, plus a
    clean production build, before it can merge.

  • Database changes are additive and reviewed, and every new data table ships
    with isolation rules turned on.

Done this way, AI-first development is actually a security advantage: changes
are consistent, documented, tested, and independently reviewed — every time,
without fatigue.

Being honest about what we don’t have (yet)

We’d rather tell you directly than have you wonder:

  • Aventide is not itself SOC 2 certified. We’re built on SOC 2-audited
    infrastructure and architected toward those same controls, but we haven’t
    completed our own SOC 2 audit. We’ll pursue it when our customers’ needs call for
    it — and we will never claim a certification we don’t hold.


If any of these is a hard requirement for your organization, talk to us — we’re
glad to walk through our architecture, our timeline, and what we can provide in the
meantime.

Found a security issue?

Email support@aventide.ai with the details. We’ll acknowledge it and keep you
posted on the fix. Please give us a reasonable chance to respond before disclosing
publicly.