AI Agent Security: A Dedicated Machine, Your Data Kept Separate

The rule that matters most

The human approval gate.

By default, nothing is sent, published, paid or deleted without a person saying go. Andy drafts, stages and proposes. A human approves. The approval rules are yours to set, and out of the box the gate sits in front of:

Sending an email or a message. Andy writes it, you say send.
Deploying a change to your website. Andy shows you the exact change first and waits for your go-ahead.
Spending money. Every purchase, subscription or paid action needs an explicit yes.
Deleting data. Nothing is deleted until you confirm it.
Touching a new system. Andy only gets the integrations you grant, with the scopes you approve.

Routine housekeeping runs on its own: labeling, sorting and archiving email, and adding, moving or updating calendar entries, including the automatic updates your calendar sends to attendees. Reading, summarizing and drafting need no sign-off either. You can tighten or loosen these rules any time, by asking Andy or in its settings, and every change to them is logged. We recommend keeping approval for payments and for anything sent outside your organization. You can see what the default looks like in practice on the how it works page: every send, payment and deletion waits for your approval before it runs.

The architecture

Eight controls, built in.

Separation

A dedicated machine, your data kept separate

Every customer runs on a dedicated cloud machine that we operate for them, and your data is kept separate from other customers’ data. Your memory, files and prompts are not pooled with anyone else’s, and customers are separated by machine, not by a checkbox in a database.

That machine is where everything about your deployment lives: the memory, the files, the credentials, the rules. Keeping your data separate isn’t a security add-on. It is how the whole product is built.

See what runs on that machine

Secrets

Restricted credentials

The access credentials you grant are stored with access restricted to Andy Labs’ systems and kept separate from every other customer’s. Data moves encrypted over TLS 1.2 or higher.

Ownership

Your data and memory are yours

You own your data, the memory built from it, your customizations, and everything Andy produces for you. You also choose how the AI is supplied: your own Claude subscription (the default), your own Anthropic API key, or AI we supply, and you can switch any time.

Access control

Least privilege by default

Andy only gets the integrations you grant, with the scopes you approve. By default, sending, posting, paying and deleting wait for explicit confirmation, and you set those rules.

Auditability

A record of what Andy did

Andy keeps a record of what it did, and every change to your approval rules is logged. Not a black box.

Data residency

Hosted in the region you choose

Memory, files and history live on the dedicated machine we run for you, in the region you choose from the ones we offer, including your own country. Data held there stays in that region. Content sent to the AI model for each task is processed where the AI provider operates, currently the United States.

Resilience

Updates without the drama

Improvements roll out with nothing for you to do. If an update affects one of your customizations, we tell you in advance where we can and help you adjust it.

No lock-in

Portable by design

Your data and memory are yours, and there is no minimum term. Ask for an export any time, not just when you leave, and you get it in common, usable formats. You can also buy out your Andy and have us move it to infrastructure in your own name. The relationship is built on the work it does, not on trapping your data.

In plain language

What that means day to day.

Your credentials are stored with access restricted to our systems, separate from every other customer’s.
By default, nothing is sent, paid or deleted until you say go. You set the rules, and every change to them is logged.
Hosted in the region you choose, including your own country. Model processing happens where the AI provider operates.
Andy keeps a record of what it did.
We never train AI models on your data, and we never sell it.
If we confirm a security breach affecting your data, you hear from us within 72 hours.
We hold no SOC 2 or ISO certification, and no system is completely secure. We take reasonable measures and tell you plainly what they are.
We don’t promise backups of the data held in the service, so ask for an export any time to keep your own copy.
We keep root and administrator control of the machine. You work with Andy through chat, email and its settings.

Built by people who handle sensitive data. Andy Labs grew out of running a real business under strict privacy law, where careless data handling isn’t an option. You can see that company running on Andy today on the proof page. Health information, payment card data and other special-category data stay out of Andy unless we have both signed an addendum for it.

The paperwork

A real entity, and a real contract.

Architecture is only half of trust. The other half is who you are actually signing with, and what that document says about your data.

Andy Labs LLC

A registered Wyoming company. You contract with a real legal entity, not a handshake and a Stripe link. More on who is behind it on the about page.

One Customer Agreement

One plain-English Customer Agreement, in writing before anything is set up: what we run for you, what happens if you stop, and who owns what. You read it before you sign it.

Your data, in writing

How your data is handled, where it lives, which providers touch it, and what happens to it when the relationship ends are all in the agreement. If you need a GDPR data processing addendum, it is available on request.

You get the full agreement before you sign anything. If your counsel wants changes, that conversation happens first.

Security FAQ

The hard questions, answered.

How is my data separated from other customers?

Every customer runs on a dedicated machine we run for them, with its own memory, files and credentials, and your data is kept separate from other customers’ data. Their data is not on the machine we run for you, and yours is not on theirs.

Can Andy send or delete things on its own?

You set the boundaries. By default, sending an email or message, posting or sharing anything outside your organization, making a payment and deleting data all wait for your approval, so Andy drafts and proposes but you decide. Routine housekeeping, like sorting and archiving email and keeping your calendar up to date, runs on its own. Andy won’t touch systems you haven’t granted. The approval rules are yours to set. You can tighten or loosen them any time, by asking Andy or in its settings, and every change to them is logged.

Am I locked into one AI vendor?

No. You choose how the AI is supplied: your own Claude subscription (the default), your own Anthropic API key, or AI we supply, and you can switch any time. The rest has no lock-in either. Your data and memory are yours, you can export them any time, or you can buy out your Andy and we move it to infrastructure in your own name.

More questions? The full FAQ lives on the get started page, and every one of them is fair game on the demo call.

A dedicated machine. Your data kept separate.