◆ Data access

What We Can See. And What We Can’t.

Agency owners ask us one question more than any other: can Zyan staff read my workspace? The honest answer depends on what kind of data it is. Here is the exact split, the promise we make for each class, and the changes that will make every promise checkable from inside your workspace.

See the three classes Security overview

◆ Three data classes◆ Zero-knowledge files◆ No training, any class

◆ Three data classes

Three Classes of Data. Three Different Promises.

One blanket promise would be either false or useless. So we split your workspace into three classes and state, for each one, what encryption applies, who holds the key, and whether anyone at Zyan can read it.

What is in it: Password-protected team documents (pay and tax paperwork) and locked Notes folders, including every note inside them.
How it is protected: Encrypted in your browser with AES-256-GCM before anything is uploaded. The key is derived from your password with PBKDF2-SHA256 and never leaves your device. Zyan stores ciphertext and nothing else.
Who holds the key: You, and only you.
What that means for you: Nobody at Zyan can open these files. Support cannot recover them. A lost password loses the file permanently, and the product says so before you set one. The Zyan AI assistant cannot read them either: locked folders are excluded from its tools by design.

AES-256-GCM, in your browser
Password and key never leave your device
No recovery path, on purpose

What is in it: OAuth tokens and API keys for the integrations you connect: Google, Meta, Slack, Stripe, ad platforms, email and texting providers, and any AI key you bring.
How it is protected: Stored in an encrypted vault, never in a plaintext column. Your workspace holds a pointer, not the secret. A credential is decrypted only inside a server function that is running one of your integrations, at the moment it runs.
Who holds the key: The vault. Decryption happens server-side, on your workspace’s behalf.
What that means for you: Credentials are not readable in database exports, backups, or any table your workspace is built on. The honest caveat: a Zyan platform operator with infrastructure access could invoke the vault. That access falls under the staff access policy below, and we never copy, display, or reuse a credential outside your own workspace’s jobs. Disconnecting an integration removes its credential from the vault.

Encrypted vault, never plaintext
Decrypted only inside your own jobs
Disconnect or rotate at any time

What is in it: CRM contacts and leads, pipelines, notes outside a locked folder, files without a password, email and texting threads, client portal content, automations, reports, and your Zyan AI conversation history.
How it is protected: Encrypted in transit with TLS and at rest with AES-256. Isolated per workspace by row-level security on every table and private, tenant-scoped realtime channels. Client seats only ever see their own portal.
Who holds the key: The platform. That is what makes search, automations, AI features, and support possible.
What that means for you: Zyan staff can technically reach this class, and we would rather say so than hide behind the word “encrypted”. Access is governed by the policy below: only for a support request you opened, an active incident, or a legal obligation. Never for browsing, sales, product research on your content, or model training.

Row-level security on every table
TLS in transit, AES-256 at rest
Staff access needs a reason, not curiosity

◆ Every class

No Training. No Selling. No Exceptions by Class.

The three classes differ in who can read the data. These commitments do not: they apply to your prompts, your files, your CRM, and your clients’ data alike.

We never train on your data.

Not on prompts, files, contacts, content, or your clients’ data. Not now, not aggregated, not anonymized, not for a future model. This is written into our privacy policy and it applies to every class above.

Prompts exist only to produce your answer.

When you use Zyan AI, your request goes to the model provider once, to generate that response. Anthropic and OpenAI are accessed through their commercial APIs, which do not use API traffic to train their models. Kimi (Moonshot AI) is opt-in, runs on a key you bring, and is governed by your own agreement with Moonshot.

Your AI history stays in your workspace.

Conversation history is stored in your workspace under the same isolation as everything else, so you can go back to it. It is not copied anywhere else, and you can delete conversations from inside the product.

Operators see metering, not content.

Zyan’s internal cost and billing views show token counts, model names, and cost per workspace. They cannot display prompts or responses: the database functions behind them do not return that data.

No selling, no advertising, no data brokers.

We do not sell workspace data, connected-platform data, or your clients’ information, and we do not use any of it for advertising. Our subprocessors process data only to run the feature you are using.

When the mechanism changes, this page changes first.

Every mechanism named here is the one that is shipped. If we add a class, move data between classes, or change who holds a key, the change lands on this page before it lands in a sales deck.

◆ Staff access policy

Who at Zyan Can Reach Standard Data, and When.

Class 03 is the class we could read. This is the rule that says when we may, written the way we would want a vendor of ours to write it.

Who

Named platform operators on an allowlist. No contractors, no third parties, and no support tier with blanket access.

When

Only for one of three reasons:

A support request you opened, limited to the workspace it concerns.
An active security or availability incident.
A legal obligation we are required to meet.

Never

Browsing. Sales research. Product research on your content. Model training. Anything you did not ask for.

How

Minimum necessary. The least data, for the least time, tied to the request that justified it. Access ends when the request does.

Your clients

A client seat can never read the agency’s workspace beyond its own portal, and your clients cannot see each other.

In the product

There is no “log in as customer” feature in Zyan. Staff cannot open your workspace from the app. The only path is infrastructure access, which is exactly what this policy governs.

Where this stands today: a written commitment, backed by the privacy policy and the DPA. The roadmap below is what turns it into a control you can inspect from inside your workspace, and this page updates as each piece lands.

◆ From promise to proof

What Is Shipped, and What Is Next.

Shipped means you can rely on it today. Next means it is planned and not yet live. We do not mark something shipped to make this page look better.

Shipped
No-training commitment in writing
In the privacy policy and on this page, for every data class. Operator views show token counts and cost, never prompts or responses.
Next
A staff access log you can read
Every staff read of your workspace writes an entry to your own activity ledger with the reason and the ticket. An empty log is the proof.
Next
Support access you grant, operator access hardened
A time-boxed “grant Zyan support access” switch in Settings, statement-level logging of direct database access, and service credentials kept off staff laptops. SOC 2 stays on the security roadmap.

The platform security roadmap, SOC 2 included →

◆ Questions

Data access questions

Can Zyan staff log in to my workspace?

No. There is no impersonation or “view as customer” feature, so staff cannot open your workspace from the app. Infrastructure access to standard data exists and is governed by the policy above.

Can you reset the password on a locked folder or a protected document?

No. The password never reaches us, so there is nothing to reset. If it is lost, the file is gone. The product warns you before you set one.

Does the Zyan AI assistant read my locked notes?

No. Locked folders are excluded from the assistant’s tools. It can see that a folder is locked, not what is inside it.

Is my clients’ data used to train AI?

No. Nothing in your workspace, including your clients’ data, is used to train models, by us or by our providers. See the commitments above and our privacy policy.

Can your team see my integration tokens?

They are stored in an encrypted vault and never shown in the product. An operator with infrastructure access could invoke the vault, which is why that access is policy-bound, and why the access log on the roadmap covers it. Disconnecting an integration removes its credential from the vault.

Where is my data hosted?

In the United States, on managed cloud infrastructure. The subprocessor list names each provider and what it is used for.

What happens to my data if I leave?

Delete your workspace or send a verified deletion request, and eligible data is deleted or de-identified from production within 30 days. Encrypted backups rotate out on their own schedule. The steps are on the data deletion page.

◆ In writing

Need This on Letterhead?

The trust packet includes the DPA, this data access statement, the incident-response runbook, and the full subprocessor table. Share it with whoever signs off on vendors.

Request the trust packet Read the questionnaire