WorkerAI

Privacy Policy

---

Effective 14 August 2026 · Last updated 14 August 2026 · Version 1.0

This policy explains what WorkerAI does with information — including the data in the email accounts, chat accounts, file storage, calendars and databases you connect to it. It is written to describe how the product actually works, not in generalities.

1. Scope of this policy

WorkerAI is a service operated by Applied AI Limited (“we”, “us”). It lets an organisation connect its data sources — databases, email accounts, messaging accounts, file storage and calendars — and work with them through an AI agent that reads across those sources, answers questions, builds reports, and (with explicit approval) writes back or sends messages.

This policy applies to the WorkerAI web application, its API, its MCP endpoints, the agent interfaces available through WhatsApp and WeChat Work, and the marketing website. It covers personal information as defined in the New Zealand Privacy Act 2020. Where we serve customers in Australia, we also handle personal information consistently with the Australian Privacy Principles.

It does not cover the separately licensed self-hosted deployment, where the organisation runs the software on its own infrastructure and is the sole holder of the data.

2. How WorkerAI handles data — in short

Three things matter more than the rest, so we state them plainly before the detail.

Connected sources are read, not copied. When you connect Gmail, Outlook, WhatsApp Business, WeChat Work, Google Drive, OneDrive, Google Calendar or Outlook Calendar, we do not import or index your content. The agent fetches what it needs from the provider’s API at the moment it answers a question, uses it, and does not retain it. Your mailbox stays the authoritative copy. Disconnecting an integration removes the connection — there is no stored content to purge.

There are real exceptions, and we list them. Some features exist precisely to keep data. Saved reports and charts store the rows they were built from. Conversations and their attachments persist so you can return to them. Agent memory keeps facts your organisation taught it. Section 6 sets out each one.

Your data is not used to train AI models. Not ours, and not the model providers’. Content is sent to a model provider to generate a response to your request, and for no other purpose.

3. Our role, and your organisation’s role

WorkerAI is sold to organisations. That splits responsibility in a way worth understanding:

• For the content your organisation puts through the agent — emails read, database rows queried, files fetched, messages drafted — your organisation decides what is connected and why. We handle that information on your organisation’s instructions, to deliver the service. Your organisation is responsible for having the right to connect each source, and for whether its use of that data is lawful and disclosed to the people concerned.
• For account, billing and usage information — your name, email, login credentials, subscription and payment records, and the metering of AI usage — we determine how that information is handled, and we are directly accountable to you for it.

Practically: if you want to know what is in your organisation’s connected sources, or want content removed from them, that is a matter for your organisation’s administrator. If your request concerns your WorkerAI user account, come to us directly.

Administrators should also understand that the agent’s access is bounded by two independent layers — a member’s role, and the specific integration instances they have been granted. Both must pass. A member with no grant for a mailbox cannot reach it through the agent regardless of their role.

4. Information we collect directly

5. Information we access from your connected sources

These integrations are read on demand. The agent calls the provider’s API while answering your request, uses the result to form its response, and does not store the content. The list below is what each connection makes reachable.

Every send, every database write and every calendar change is shown to you as a preview and executed only after you approve it. All of them are recorded in the audit log.

6. Information from your sources that we do store

Section 5 describes the default. These are the deliberate exceptions, and we would rather be exact about them than leave you to discover them.

7. AI processing and model providers

Answering a query means sending content to a large language model. What is sent is your prompt, the conversation so far, the definitions of your connected sources, relevant agent memory, and whatever the agent retrieved — email excerpts, file contents, query results.

Whose key is used changes who you are dealing with.

• Platform keys. If you use WorkerAI’s AI capacity, calls go to model providers under our commercial agreements with them, on terms that prohibit using submitted content to train their models, and subject to those providers’ usage policies. Usage counts against your organisation’s monthly allowance.
• Your own keys. If your organisation configures its own provider API key, calls go to that provider under your agreement with them, and their terms govern that content. We still record usage for transparency, but it does not count against your allowance.

Model providers we may use with platform keys: OpenAI, Google (Gemini) and Anthropic. Custom OpenAI-compatible endpoints can also be configured by your organisation, in which case that endpoint’s operator receives the content.

Related processing to be aware of:

• Prompt caching. To reduce cost and latency, portions of a prompt may be cached with the provider for a short period. Cache partitions are separated per organisation.
• Web search. When a query cannot be answered from your sources and web search is enabled, a search query — not your source data — is sent to our search provider.
• Tool and MCP calls. The agent passes data to the tools and external MCP servers your organisation has connected.
• Optional redaction. The platform can detect and mask common identifiers, such as phone numbers, in stored traces. This is a configurable control and is not enabled by default — do not treat it as a guarantee that stored traces are free of personal information.

8. How we use information

• To provide the service: authenticate you, route queries, retrieve context, generate responses, execute approved actions, and run scheduled work.
• To operate access control and keep organisations isolated from one another.
• To measure usage against your plan’s monthly allowance, bill you, and show cost transparency.
• To secure the platform: detect abuse, investigate incidents, and maintain audit trails.
• To support you when you ask for help, and to send service and transactional messages such as verification, invitations, billing notices and task-completion notifications.
• To improve the service — using aggregated and operational data such as error rates, latency, model performance and failure patterns. We do not use your content to train AI models.
• To meet legal obligations, and to establish or defend legal claims.

9. Who we disclose information to

We do not sell personal information, and we do not disclose it for advertising. We use the following service providers to run WorkerAI:

We also disclose information:

• To providers your organisation connects — Google, Microsoft, Meta, WeChat Work, your own database hosts, and any external MCP server your organisation adds. Those services are governed by their own terms and privacy policies. Choosing to connect an MCP server means choosing to send query data to whoever operates it; we cannot vouch for third-party servers.
• Within your organisation — to administrators and to members you share with, as described in section 12.
• Where the law requires it, or to protect the rights and safety of people or of Applied AI Limited.
• In a business sale or restructure, subject to the acquirer continuing to handle the information under terms no less protective than these.

We will keep this list current. Material additions will be reflected here with an updated version date.

10. Overseas storage and processing

WorkerAI is operated from New Zealand, but the service depends on providers located overseas — principally in the United States, the European Union and Australia. Your information, including content retrieved from your connected sources, will be processed outside New Zealand.

We only use providers who are subject to privacy obligations comparable to those in the New Zealand Privacy Act 2020, or who are bound by contractual terms requiring comparable protection. By using WorkerAI you accept that your information is handled overseas on this basis.

11. Security

What we do:

• Credentials are encrypted at rest. OAuth tokens and API keys are encrypted before being written to the database. After you save an API key, only its last four characters are ever displayed.
• Two independent authorisation layers. A member’s role governs what kinds of action they may take; per-integration access grants govern which specific sources they may touch. Both must pass. The agent’s reach is capped by the grants of the person asking — a member cannot use the agent to see something they could not see directly.
• Row- and column-level rules. Administrators can define conditional access rules that constrain which rows and columns a member’s queries return. These are enforced on the server, including on agent paths.
• Session-level organisation isolation. Each session is bound to one organisation; requests carrying a mismatched organisation are rejected.
• Two-factor authentication, by authenticator app or emailed one-time code. Administrators can require it for every member of an organisation.
• Confirmation before consequence. No database write, no outbound message and no calendar change happens without a human approving a preview of it.
• Append-only audit logging of significant actions, and private object storage for attachments.
• Encryption in transit, rate limiting, and standard security headers.

Two limits we would rather state than let you assume:

• Isolation between organisations is enforced in the application layer — through scoped queries and authorisation checks — not by database-level row security. It is thorough and tested, but it is not a database-enforced boundary.
• Redaction of personal information in stored traces is optional and off by default, as noted in section 7.

No system is perfectly secure. If a privacy breach occurs that is likely to cause serious harm, we will notify the Office of the Privacy Commissioner and affected people as the Privacy Act 2020 requires.

12. Features that make data visible to others

Several features exist to share results. Each one widens who can see data, so each is described here.

• Conversations are private by default. Only the person who created a conversation can see it. If you share one, the members you share it with get read-only access; if they edit it, they get their own private copy rather than changing yours.
• Public view links. A saved view can be published as a link containing an access key. Anyone holding that link can see the data in that view without signing in. The link is scoped to that view’s stored data only, and the page is read-only, but treat the link itself as the credential — anyone you send it to can forward it. Revoke it when it is no longer needed.
• Chat over a published view. Where enabled, viewers can ask questions about a published view. Those questions are answered strictly from that view’s stored data; the agent cannot reach your live sources from there.
• MCP server exposure. An administrator can expose selected workflows and resources to external AI clients through our MCP endpoint. Those clients operate within the same role and grant model, and every invocation is audit-logged.
• Organisation API keys and outbound webhooks. API keys allow programmatic access under the permissions assigned to them. Webhooks send event data to a URL you nominate. Both are ways data leaves the platform under your organisation’s control.
• Messaging interfaces. Members can use the agent through WhatsApp or WeChat Work after linking their number or account. Their role and grants apply exactly as on the web.

13. Retention and deletion

Most data in WorkerAI is kept until you delete it, or until the organisation is deleted. We do not silently expire your work. The exceptions below are the only automatic deletions in the system.

When a member leaves an organisation, their access is revoked immediately and their private conversations become inaccessible to them. Context they contributed — source descriptions, rules, corrections and agent memory — stays with the organisation and remains active, because that knowledge was contributed on the organisation’s behalf. It may be tagged as contributed by a former member for audit purposes.

When an organisation closes its account, we delete its data within 30 days, apart from records we are required to keep as set out above. Export what you need before closing.

14. Your rights — access and correction

Under the Privacy Act 2020 you may ask for access to the personal information we hold about you, and ask us to correct it. Australian customers have equivalent rights under the Australian Privacy Principles.

Much of this you can do yourself: update your profile, change your password, manage two-factor authentication, review connected sign-in accounts and organisation memberships, delete conversations, leave an organisation, or delete your account. For anything else, contact info@workerai.co.nz.

We will respond within 20 working days. We may need to verify your identity first. If we decline a request — for example where releasing information would reveal someone else’s personal information — we will explain why, and you may complain to the Privacy Commissioner.

Where your request concerns content inside an organisation’s connected sources, we will direct you to that organisation’s administrator, who controls what is connected and what is retained. We will assist them in responding.

15. Cookies and browser storage

We use no advertising cookies and no third-party tracking cookies. What we do use:

• A session cookie that keeps you signed in. Essential.
• A small cookie recording which second-factor method to present at sign-in, so the login screen can show the right challenge. Essential.
• Local storage for interface preferences such as your theme and sidebar state. This stays in your browser.

Blocking essential cookies will prevent you signing in.

16. Children

WorkerAI is a business tool and is not directed at children. We do not knowingly collect personal information from anyone under 16. If you believe a child has provided us with personal information, contact us and we will delete it.

17. Changes to this policy

We may update this policy as the product changes. The version and dates at the top always reflect the current text. Where a change materially affects how we handle your information, we will notify account administrators by email or in-app notice before it takes effect. Continuing to use WorkerAI after that date means you accept the updated policy.

18. Complaints

Raise a concern with us first at info@workerai.co.nz. We will acknowledge it promptly and aim to resolve it within 20 working days.

If you are not satisfied, you may complain to:

• New Zealand — Office of the Privacy Commissioner, privacy.org.nz, 0800 803 909.
• Australia — Office of the Australian Information Commissioner, oaic.gov.au, 1300 363 992.

Appendix A — Google API Services Limited Use

WorkerAI’s use of information received from Google APIs adheres to the Google API Services User Data Policy, including the Limited Use requirements.

Specifically, data obtained through Google APIs is:

• used only to provide or improve user-facing features that are prominent in the WorkerAI interface, and not transferred or used for any other purpose;
• not used to serve advertising of any kind, including retargeting or personalised advertising;
• not sold to any party;
• not used to develop, improve or train generalised artificial intelligence or machine learning models. Content is transmitted to model providers only to generate a response to the user’s own request, under terms that prohibit training on that content;
• not read by humans, unless we have your explicit consent for a specific message or file, it is necessary for security purposes such as investigating abuse, it is required to comply with applicable law, or the data has been aggregated and anonymised for internal operations;
• not transferred to others except as necessary to provide or improve those features, to comply with applicable law, or as part of a merger or acquisition with the user notified.

The Google permissions we request, and why:

You can review and revoke WorkerAI’s access at any time at myaccount.google.com/permissions, or by disconnecting the integration in WorkerAI.

Appendix B — Microsoft Graph permissions

The same principles apply to data obtained through Microsoft Graph: it is used only to deliver the features described here, is never used for advertising, is never sold, and is never used to train generalised AI models.

Where both Outlook email and OneDrive are connected, they may share a single Microsoft sign-in. Access can be revoked at myaccount.microsoft.com or by disconnecting the integration.