Privacy Policy

In this policy, "we", "us" and "Cerberius" mean Cerberius Pty Ltd. "Personal information" means information about an identified person, or a person who can reasonably be identified.

This policy covers:

• the Cerberius Agents website at cerberius.com, including sign-up, the plans form, the website chat, and the early access and contact forms;
• the Cerberius Agents app, its API, its AI agent, its chatbots, setup assistants and content writer, the chat widget and the phone app (together, the "service").

It doesn't cover the websites and systems of our customers, or of the third-party services they connect, which have their own policies.

Our customers are organisations. They use the service to handle support for their own customers. So personal information reaches us in two ways, and our role is different for each.

• Our own information. What you send through our website, and the details of the people who use the app. We decide how it is used, and this policy fully applies.
• Information we handle for our customers. Support tickets, chatbot conversations and what the agent reads in their systems. The customer decides what goes in and why. We handle it on their behalf and on their instructions. If you contacted one of our customers and want to ask about your information, please contact that organisation first. If they can't help, write to us and we'll help where we can.

When you send the early access form or the contact form, we collect:

• your name, email address and company (company is optional on the contact form);
• anything else you choose to give us: your website, role, team size, ticket system, the features you're interested in, the topic and your message;
• for early access requests, where the request came from: the page you sent it from, the site that linked you here (if your browser shares it) and any campaign tag in the address;
• your browser's user agent (the name and version your browser reports);
• a keyed hash of your IP address. This is a scrambled code, made with a secret key, that lets us count how many forms come from one address. We don't store your IP address itself;
• how long the form was open before you sent it, to spot automated sign-ups. This isn't stored.

To keep out automated spam, your browser also solves a small computing puzzle while you fill in the form. It runs on your device, takes about a second, and involves no third party. We keep only the solved puzzle, for 20 minutes, so it can't be used twice. Nothing about your device is stored beyond what's listed above.

We use this to reply to you, to decide on and set up early access, to talk with you about Cerberius Agents, and to stop spam and abuse. Each form is stored in our system and emailed to our sales team. Our team also gets a short notice in its own Slack or Telegram chat, with your name, company, email address and the start of your message. You don't receive an automatic email.

You don't have to give us any of this. Without your name, email and (for early access) your company, we can't respond to your form.

Signing up

When you sign up, we collect your name, your organisation's name, your email address (from Google, Apple or the address you type for a code), your browser's time zone, the time you accepted our terms, and a keyed hash of your IP address to limit how many sign-ups one address can start. An unfinished sign-up is deleted after an hour. A finished one becomes your organisation and your membership of it (section 5). Your time zone is passed to our billing service with your organisation's billing account, and sets the currency you're billed in.

Choosing a paid plan

The plans form on our pricing page is provided by our billing service, NicelyBilled. In it you give your name and email address, and your organisation if you like. Your card or PayPal details go straight to the payment processor. Your IP address is used to work out which country you're in, for the currency. NicelyBilled sends us the billing contact's name, email address and organisation, and the plan you chose, so we can set up your organisation and email you a link to sign in.

The website chat

The chat on this website and in the help centre is one of our own chatbots. It answers from our help centre. We keep your questions and its answers, a random id your browser keeps for the conversation, and a keyed hash of your IP address to limit message rates. Your questions are sent to the AI provider the chat uses (Anthropic or OpenAI), under our own account (section 9). Please don't type passwords, card numbers or other sensitive details into it.

• No tracking. The website has no analytics, advertising or social media trackers, and sets no cookies for those purposes.
• Sign-in cookie. When you sign in to the app, we set a cookie that keeps you signed in. It is needed for the app to work, and scripts on the page can't read it.
• Browser storage. The app remembers which platform you last chose, in your browser's own storage. On this website and on our customers' websites, the chat widget keeps a random visitor id and the recent conversation in the visitor's browser, so the chat carries on between pages. The visitor can clear it by clearing their browser's site data.
• The plans form. The plans form on our pricing page, and the billing pages in the app, show our billing service's form in a frame. It and the payment processor's own scripts (Square, Stripe or PayPal) may set cookies or use browser storage of their own, to take payments and prevent fraud. Their own policies apply to those.

For people who are members of an organisation in the app, we collect:

• your name and email address, from your Google or Apple account when you sign in, the address you sign in with by email code, and the email address your organisation added for you. If you choose Apple's Hide My Email, we receive Apple's private relay address instead of your own;
• your organisation and your role in it;
• if your organisation has a plan, its billing details: the organisation's name, the billing contact's name and email, the plan, invoices and payments, and how much of the plan is in use (tickets the agent started on, and the number of platforms, people and chatbots). Card and PayPal details go straight to the payment processor; we don't receive or store them;
• a record of what you do in the app, such as replies you approve, edit or skip, fixes and changes you approve, fixes you merge, messages to the agent, Development requests and their attachments, feedback, and settings you change;
• if you use the help chat inside the app: your questions, its answers, and a short summary of your organisation and the platform on screen (such as its name and what is connected, never keys or ticket content), so it can answer about your setup. The help chat is one of our own chatbots;
• emails and other messages you send us, such as support requests.

We use Google and Apple only to confirm who you are. We don't receive your password, and we don't access your Google or Apple account's other data.

The Cerberius Agents phone app lets members see what waits for them and act on it. You sign in through this website, in your phone's browser, and the app then gets its own sign-in key.

• On your phone, the app keeps the sign-in key (in the phone's secure storage), the address of the service, which platform you last chose, and whether the app lock is on. Nothing else is stored on the phone, apart from what is on screen while you use it.
• On our service, we record the phone's name and type (for example "Pixel 10, Android"), its push notification address, which notifications you chose, and when it was last used.
• Notifications carry the platform's name, the ticket's subject or the article's title, a short line such as who sent it, and the internal ids the app needs to open the right screen. They don't carry the reply or the agent's findings.
• Writing to the agent. What you write to the agent in the app, such as on a ticket or in a Development request, is handled like the same message in the web app: stored with the ticket and sent to the AI provider under your organisation's key (section 9).
• App lock. If you turn on the lock, your phone checks your fingerprint, face or PIN itself. We never receive any of it; the app only learns whether the check passed.
• No tracking. The app has no analytics, advertising or tracking tools.

Signing out in the app, or uninstalling it, stops its notifications. Signing out also makes its sign-in key stop working straight away.

Depending on what a customer connects and switches on, the service handles:

• Support tickets that the customer's ticket system sends, or that we read from the customer's HaloPSA, with follow-ups and attachments. We keep a copy of each attached file. Tickets imported from HaloPSA bring their whole conversation, including the customer's team's replies and private notes. Tickets can also come from errors in the customer's logs. They often contain personal information about the customer's own customers, clients or staff, such as names, email addresses, phone numbers, account and booking details, and whatever the person wrote.
• What the agent reads while investigating: the customer's code and build results, logs, Kubernetes details, the results of read-only database queries, Microsoft 365 and Google Workspace details (such as users, sign-ins, groups, licences, mailbox settings and devices) and DNS records. If the customer gives the agent its own machine, it also includes the results of the checks the agent runs there. These can include personal information about the customer's staff, clients and customers.
• What the service produces: the agent's findings, the steps it took, the commands run on the agent's own machine and what came back, its working record (transcript) for each ticket, drafts and replies, proposed fixes and changes, and their results.
• Chatbot conversations: visitors' questions and the chatbot's answers, the website the chat came from, a random visitor id and a keyed hash of the visitor's IP address (to limit message rates). For signed-in visitors, also the account id and name the customer's own website gives us.
• Setup assistant conversations: when a customer puts a setup assistant in its own portal, the messages of the customer's signed-in users, the account id and name the customer's server gives us for them, what the assistant read and proposed, what they approved, and the results. The short-lived key the customer's server gives the assistant for each user is used only for that conversation and isn't stored.
• Development requests: what the customer's developers ask for, the pictures and files they attach, the agent's replies, and the pull requests that result.
• Content writer: the market research, topic ideas and article drafts it writes for the customer. Its research reads public web pages only.
• Keys and logins the customer adds. These are encrypted, and are not used to identify anyone.

For this information, the customer is responsible for having a lawful basis to collect it and to have it handled by us and by the AI provider they choose, and for giving their own users the notices the law requires. We use it only to provide the service to that customer, to support and secure it, and as the law requires. We don't sell it, and we don't use it for our own marketing.

Our staff look at customer data only when needed to provide, support or secure the service, to investigate a problem the customer raised or misuse of the service, or where the law requires.

We use personal information to:

• provide the service, including investigating tickets, drafting replies and running chatbots;
• reply to your enquiries, consider early access requests and set up organisations;
• sign people in, and keep accounts and the service secure;
• bill organisations on a paid plan, apply their plan's limits, and tell our billing service how much of the plan is in use;
• prevent, detect and investigate spam, fraud, misuse and security incidents;
• provide support, and find and fix problems with the service;
• tell customers about their account, changes to the service and these policies;
• talk to business contacts about Cerberius Agents. You can ask us to stop at any time. We don't use our customers' end users' information for this;
• meet our legal obligations, and establish, exercise or defend legal claims.

We do not sell personal information. We don't use it for advertising, and we don't use it for any purpose unrelated to those above unless you agree or the law allows.

If the GDPR or UK GDPR applies to you, our legal bases are: providing the service under our contract with your organisation; our legitimate interests in running, securing and improving the service, preventing abuse, and talking to business contacts about it; your consent, where we ask for it; and our legal obligations.

Any marketing email from us says who it's from and has a way to unsubscribe. We act on an unsubscribe within 5 working days. Emails about your account, such as sign-in codes, invoices and notices, aren't marketing and keep coming while you use the service.

• To investigate a ticket, the agent sends the ticket and what it reads (code, logs, query results, account details and DNS records) to the AI provider (Anthropic, for Claude), using the AI key the customer's organisation added.
• Chatbots send visitors' questions and the customer's help content to the AI provider the customer chose for that chatbot (Anthropic or OpenAI), using the customer's key.
• Setup assistants and the content writer send their conversations, research and drafts to Anthropic, using the customer's key. The content writer's research searches the web and reads public pages. It is given only the customer's public details, such as its name, website and audience.
• That use is under the customer's own account with the provider, and the provider's terms and privacy policy apply to it. Please refer to them for how the provider handles and keeps data. If we have agreed in writing to provide a key for a customer, the information goes to the provider under our account instead.
• Our own chatbots, the chat on this website and the help chat inside the app, send your questions to an AI provider (Anthropic or OpenAI) under our own account.
• We don't use customer data to train AI models, and we don't ask providers to.
• Values that look like secrets are hidden from cluster, log and database results before the agent sees them. Ordinary personal details, such as names and email addresses, are not hidden, because investigations need them.

Some decisions in the service are made by software, without a person deciding each one:

• Plans and billing. Using your organisation's plan, its usage and whether its bills are paid, the service stops starting new tickets when the plan's allowance is used or the billing account is locked for an unpaid bill, and refuses to add platforms, people or chatbots beyond the plan. The plan, the usage and the billing contact's details are used.
• Spam and abuse. Using a keyed hash of an IP address, how often it has been used, and the small puzzle your browser solves, the service may refuse a form, a sign-up, a sign-in code or a chat message that looks automated or too frequent.
• For our customers, on their settings: automatic replies, when a customer switches them on, send some replies without a person approving each one; the content check that holds replies back; the log watcher's daily limit; and the per-ticket cost limit that stops an investigation. Our customers decide whether to use these, and tell their own users.

Otherwise the agent only proposes. A person decides what is sent and what is changed, apart from changes a customer's own user approves in a setup assistant, and Development changes someone chose to merge and deploy without asking. To ask about a decision, or have a person look at it, email [email protected].

We share personal information only as this policy describes:

• Our hosting providers: HostHatch, whose data centre in Sydney, Australia runs our servers, and DigitalOcean, which stores our backups in Sydney.
• AI providers: Anthropic, and OpenAI where a customer chooses it for a chatbot, and for our own chatbots, as described in section 9.
• Code hosts: GitHub and GitLab, when a customer connects repositories. Approved fixes are sent to them as pull or merge requests.
• HaloPSA, when a customer connects it. We read tickets and their conversations from the customer's Halo, and send back approved replies, the agent's findings as private notes, and the assignment, status and close changes the customer chose.
• Microsoft and Google, when a customer connects Microsoft 365 or Google Workspace. We read from them for investigations, and send them the changes a person approved.
• GoDaddy, when a customer connects it, to read DNS records and make record changes a person approved.
• Public internet servers, only if a customer gives the agent its own machine and lets it reach the internet. The checks it runs there, such as looking up a domain's records or testing a mail server, contact public servers.
• Google and Apple, to sign people in.
• Our billing service, NicelyBilled, and its payment processors, such as Square, Stripe and PayPal, to bill organisations: the organisation's name, the billing contact, the plan, invoices, payments, and how much of the plan is in use.
• Expo, Apple and Google, to deliver the phone app's notifications. We send each notification to Expo's push service, which passes it to Apple (for iPhones) or Google (for Android phones), who deliver it to the phone. Section 6 lists what a notification contains.
• Slack and Telegram, when a customer sets up notifications. Notifications carry only the platform, the ticket's subject, who sent it and a link. We also use them for short notices to our own team about website forms.
• The customer's own systems, such as its ticket system (approved replies, sent to its signed webhook), its databases, its Kubernetes cluster, its Microsoft 365 or Google Workspace, and its domains at GoDaddy (approved changes).
• Our email server, which sends sign-in codes, invitations, welcome emails and notices, and delivers website forms to our sales team.
• Our professional advisers, such as lawyers and accountants, under a duty of confidence.
• Law enforcement, regulators and courts, where the law requires or allows it, or to protect the rights, property or safety of anyone.
• A buyer or successor, if all or part of our business is sold, merged or reorganised. They would have to handle it in line with this policy.

Customers decide which of the third-party services above are used for their data.

We store the service's data in Australia: our servers are in Sydney, and so are our backups.

Some of the services above store or process information outside Australia. The AI providers (Anthropic and OpenAI), GitHub, GitLab, Google, Microsoft, GoDaddy, Slack, Expo, Stripe and PayPal mainly process it in the United States, and may also use other countries. HaloPSA and Telegram, and the payment processors, may process it in the United States, the United Kingdom, the European Union or elsewhere, depending on the service and the customer's own account. Which ones a customer's data reaches depends on what the customer connects and switches on.

When we disclose personal information overseas, we take reasonable steps, as required by Australian Privacy Principle 8, to ensure it is handled consistently with the Australian Privacy Principles. Where a customer connects a service under its own account, that disclosure is made on the customer's instructions and under the customer's agreement with that service.

We take reasonable steps to protect personal information from misuse, interference, loss, and unauthorised access, change or disclosure. These include:

• encrypting stored keys, tokens and passwords. Each is tied to its organisation and can't be read back once saved;
• hiding values that look like secrets from tool output, error messages and logs before they are stored;
• keeping each organisation's data separate, and limiting each member's access by role;
• keeping the agent read-only during investigations. It has no shell and no web browser. It reaches the internet only through its own machine, if a customer switches that on, and that machine holds none of the customer's keys. The content writer's research is the only other part that reads the web, and it is given no private information;
• hashing visitors' IP addresses rather than storing them.

More detail is on our Security page. No system is completely secure, and we can't guarantee the security of information. We don't hold any security certification.

If we become aware of a data breach, we'll act quickly to contain it and assess it. We'll comply with the Notifiable Data Breaches scheme in the Privacy Act. That means telling the people affected and the Office of the Australian Information Commissioner when a breach is likely to cause serious harm. If a breach affects information we handle for a customer, we'll tell that customer so it can meet its own obligations.

• Website forms are kept while they're useful for following up with you and for our business records. There's no automatic deletion. Ask us and we'll delete yours, unless we need to keep it by law.
• Customer data is kept while the customer uses the service. Deleting a platform permanently removes its tickets, including their attached files, history, transcripts, settings and chatbots. There's no automatic clean-up of old tickets yet, so the agent's transcripts and chatbot conversations are kept until the platform is deleted or the customer asks us to delete them.
• Code is downloaded fresh for each run and isn't kept afterwards.
• Sign-in details: an unfinished sign-up is deleted after an hour, and a sign-in code works once, for a few minutes. Signing in on the web lasts up to 7 days. A phone stays signed in until you sign out in the app or remove the phone.
• Website chat conversations are kept while they help us improve our help centre, and deleted on request.
• Backups. We back up the service's database twice a day, in Sydney. We don't delete old backups yet, so something deleted from the service stays in the backups taken before it was deleted. Backups are used only to restore the service after a failure.
• After a customer leaves: when an owner deletes the organisation, its data is deleted straight away. If its access ends another way, we keep its data for 30 days so it can ask for a copy, then delete it. Either way, we keep what the law requires.
• Member details are kept while the person is a member, and in the records of what they did for as long as those records are kept.

When we no longer need personal information, we delete it or remove what identifies people.

You can ask us for access to the personal information we hold about you, and ask us to correct it. Email [email protected]. We'll respond within a reasonable time, usually within 30 days. We may need to confirm who you are first. If we refuse, we'll tell you why, unless the law says we needn't.

You can also ask us to delete your information, or to stop contacting you about Cerberius Agents. Delete your data explains how, and what we delete or keep.

If you're outside Australia, the law where you live may give you further rights, such as to object to or restrict how we use your information, or to receive a copy in a portable form. Email us and we'll respond in line with the law that applies.

If your request is about information we handle for one of our customers (such as a support ticket you sent them), we'll pass it to that customer or ask you to contact them, as they decide what happens to it.

If you have a concern about how we've handled your personal information, email [email protected] with the details. We'll acknowledge it, look into it, and reply in writing within a reasonable time, usually within 30 days.

If you're not satisfied with our response, you can complain to the Office of the Australian Information Commissioner (OAIC) at www.oaic.gov.au. If you're in the European Union or the United Kingdom, you can also complain to your local data protection authority, or in the UK to the Information Commissioner's Office.

The website and the app are for businesses and are not directed at anyone under 18. We don't knowingly collect personal information from children through them. Our customers decide what their chatbots and ticket systems collect, and are responsible for it.

We'll update this policy when what we do changes, and publish the new version here with a new "Last updated" date. If a change materially affects how we handle customers' information, we'll tell them by email or in the app before it takes effect.

For privacy questions, requests and complaints, contact:

Cerberius Pty Ltd (ABN 57 686 421 152)
[email protected]

Our Terms of Service also apply to the service.