FAQ
Questions and straight answers
What Cerberius Agents does, what it can reach, and what always stays with your team. For step-by-step guides, see the help centre.
The basics
What is Cerberius Agents?
An AI support agent for the systems you look after. When a support ticket comes in, the agent looks into it. It reads what you connect: code on GitHub or GitLab, Kubernetes, logs in Loki, Elasticsearch, OpenSearch, Datadog, Better Stack, ClickHouse, Graylog, CloudWatch, Azure Monitor or Google Cloud Logging, errors in Sentry, PostgreSQL, MySQL, MariaDB, Microsoft SQL Server and MongoDB databases, Microsoft 365, Google Workspace and DNS records at GoDaddy. Then it writes up what it found and drafts a reply. A person on your team checks the draft and decides what happens next.
Who is it for?
Anyone who answers support tickets about systems they look after. That might be a software company supporting its own product, an IT service provider looking after its clients' Microsoft 365, Google Workspace and domains, or an in-house IT team. Each product or client is its own platform. It's set up by an owner or admin, and used every day by whoever reviews replies and approves changes.
How long does an investigation take?
Usually a few minutes per ticket. Each platform sets its own limits on minutes per run, steps per run and cost per ticket, and the agent stops when it reaches one.
What's the difference between a chatbot and the support agent?
A chatbot sits on your website or customer portal and answers visitors straight away, only from help pages you choose. It can't see your code, logs, databases or anyone's account (apart from a short summary your system sends for a signed-in customer). The support agent works on tickets: it investigates your systems read-only, and a person approves its replies.
What is the log watcher?
A setting that checks a platform's logs (in Loki, Elasticsearch, OpenSearch, Datadog, Better Stack, ClickHouse, Graylog, CloudWatch, Azure Monitor or Google Cloud Logging), and its Sentry, every 5 minutes. Each new, distinct error becomes one ticket, and the agent investigates it straight away, up to a daily limit you set. Repeats of the same error add to the existing ticket. It never replies to anyone outside your team.
Access and safety
What does it need access to?
A Claude key for the agent, and at least one thing for it to look at: code on GitHub or GitLab, a Kubernetes cluster or your logs (Loki, Elasticsearch, OpenSearch, Datadog, Better Stack, ClickHouse, Graylog, CloudWatch, Azure Monitor or Google Cloud Logging), a PostgreSQL, MySQL, MariaDB, Microsoft SQL Server or MongoDB database, Microsoft 365 or Google Workspace. GoDaddy, for DNS records, can come on top. Give each one a read-only login where you can. The more you connect, the better the answers. Your cluster, logs and databases must be reachable on a public address.
Does it change anything on its own?
No. Investigations only read. The agent has no shell and no web browser. Code fixes and database, Kubernetes, Microsoft 365, Google Workspace and DNS changes are each off until an owner or admin switches them on. Then every one waits for a person, who sees exactly what will run and can edit it first. The service makes the change, not the agent. The agent's own machine, if you switch it on, runs checks without approval, but it holds none of your keys and can't reach your systems.
Will it reply to my customers by itself?
Not unless you switch on automatic replies for a platform. By default, every reply waits for a person. Even with automatic replies on, replies with a proposed fix or change, replies with open questions, and anything long or technical-looking always wait for a person. That last check is a fixed rule on the reply text, not AI.
Will it merge code?
Only when a person on your team asks. If you approve a fix, the agent edits one repository and the service opens a pull request (a merge request on GitLab). Your team reviews and tests it. It's merged only when an approver presses Merge on the ticket or asks for it in the conversation, and the service does it, never the agent. The agent doesn't run code or tests, and a fix can never change CI or git settings such as the .github folder or .gitlab-ci.yml.
Can a ticket trick the agent into doing something?
The agent treats a ticket as information to investigate, never as instructions. Our safety rules come first and your own rules can't override them. What the agent can reach is limited by read-only tools and the logins you give it, and anything that changes your systems needs a person's approval.
How are our keys and passwords protected?
Every key, token and password is encrypted when you save it (AES-256-GCM, with its own key for each value) and can't be read back, in the app or through the API. Logins with write access are used only by the service, for changes a person approved. The agent never receives them. For Microsoft 365 and Google Workspace, the agent gets a pass that lasts about an hour, used only by tools that read. It never sees the secret or the key file.
What data do you keep, and where does it go?
Cerberius Agents keeps each platform's tickets, drafts and replies, the agent's findings and steps, change records and chatbot conversations, until you delete the platform. Your code is downloaded fresh for each run and isn't kept. Values that look like secrets are hidden before anything is stored. To investigate, the agent sends the ticket and what it reads to Claude, using your own Claude key.
Setup and connections
Which ticket systems does it work with?
Service desks and help desks: HaloPSA, ConnectWise PSA, Autotask PSA, Zendesk, Freshdesk, Jira Service Management, ServiceNow and Zoho Desk. Issue trackers: Jira, Linear and YouTrack. Each connects with a key or login from that system, with no code to write. Any other ticket system can connect through our ticket API: it sends new tickets, follow-ups and closes, and approved replies come back to your webhook, signed to the Standard Webhooks specification. A developer on your side does that. Each platform connects one ticket system at a time.
Does it work with HaloPSA?
Yes. Connect Halo with a key, no code needed. Each platform takes the tickets of the Halo clients you tick for it. Approved replies go back into Halo, emailed to the requester or added as a private note, and the agent's findings can go on as a private note. Tickets opened before you connected can be imported with their whole conversation.
We're an IT provider on ConnectWise or Autotask. Does it work?
Yes. Connect ConnectWise PSA with an API member's keys, or Autotask PSA with an API user. We then set up ConnectWise or Autotask to tell us about new tickets. Make one platform per client, connect each to the same ConnectWise or Autotask, and tick that client's company. In ConnectWise, an approved reply goes back as a discussion note emailed to the contact, or as an internal note. In Autotask, it's a note the contact sees in the client portal (one workflow rule emails it too), or an internal note. Findings can go on as an internal note.
Does it work with Zendesk?
Yes. Connect with an OAuth client from Zendesk, then add a webhook and a trigger in Zendesk. Our settings page gives you what to paste. New tickets and the requester's public replies come in. Approved replies go back as a public reply or a private note, and the agent's findings can go on as a private note. Each platform can take only the Zendesk organisations you tick.
Does it work with Freshdesk?
Yes. Connect with an agent's API key, then add two automation rules in Freshdesk. Those rules need Freshdesk's Growth plan or above. New tickets and the requester's replies come in. Approved replies go back as a reply Freshdesk emails, or as a private note, and the agent's findings can go on as a private note.
Does it work with ServiceNow?
Yes, with incidents. Connect with an integration user or an OAuth client, then add one business rule in ServiceNow. We give you its script to paste. New incidents and the caller's additional comments come in. Approved replies go back as an additional comment the caller sees, or as a work note for your team, and the agent's findings can go on as a work note.
Does it work with Jira?
Yes, Jira Cloud, in two ways. With Jira Service Management, requests from your service desks come in, and approved replies go to the customer as a public comment, or stay as an internal note. With Jira, issues in your team's own projects come in, and replies go on the issue as comments, which everyone who can see the issue sees. Both connect with an Atlassian API token and a webhook in Jira.
Can it work on tickets we already have?
Yes, from HaloPSA. Import tickets brings in tickets opened before you connected, with their whole conversation. Nothing is sent and nothing changes in Halo. You choose which ones the agent starts on now. The rest wait as Not started until someone starts them. From other ticket systems, tickets come in from the moment you connect.
Can it look into Microsoft 365 or Google Workspace?
Yes. Connect a Microsoft 365 app registration or a Google Workspace service account to a platform. The agent can then read users, sign-ins, groups, licences and devices. It can propose changes, such as unblocking a sign-in or adding someone to a group, and each one waits for a person. It never changes apps, admin roles or policies.
Can it check and fix DNS records?
Yes, for domains at GoDaddy. List the domains a platform may use. The agent reads their records and can propose exact records to add or change, such as the ones email needs. Each change waits for a person. Name servers are never changed.
Can the agent run its own checks, like looking up a domain's mail records?
Yes, if you give it its own machine. It's off until an owner or admin switches it on for a platform. It's a separate computer for each ticket, with none of your keys, and every command shows on the ticket. It can reach public websites and servers only if you allow the internet too.
Do we need a ticket system?
No. A platform can get its tickets only from the log watcher. Then approving the agent's findings records your decision and marks the ticket done, and nothing is sent anywhere.
Does it work with GitHub and GitLab?
Yes. GitHub with a fine-grained personal access token, and GitLab (gitlab.com or your own server) with a personal, project or group access token. Read-only access is enough to investigate. Write access is only needed on repositories where you want pull requests for approved fixes.
Which languages and frameworks does it support?
The agent searches and reads the files in your repositories, so it isn't limited to a list of languages or frameworks. It does better when you tell it how your system fits together, in the platform's "About your system" notes and each repository's description.
How long does setup take?
It depends on how much you connect. The basics are a Claude key, one thing for the agent to look at (such as your code, or a client's Microsoft 365) and a way for tickets to arrive. Everything else can come later. The help centre has a step-by-step checklist, and the app shows what's left to do.
Can we approve replies from our own tools?
Yes. An API key with the "own approval screen" permission can list waiting drafts, read the agent's analysis, approve (optionally edited) or reject, and talk to the agent. Version checks stop two people acting on the same draft twice.
We run several products, or look after several clients. Does that work?
Yes. Make one platform per product or per client. Each has its own tickets, connections, chatbots and settings, and the agent working on a ticket sees only that platform's systems. Your organisation holds members, keys, rules and notification channels, and the Overview shows every platform at once. With a shared ticket system, such as HaloPSA, ConnectWise PSA or Zendesk, each platform takes only the clients, companies or organisations you tick for it.
AI, pricing and access
Which AI does it use?
The support agent runs on Claude, using a Claude key you add. Chatbots can use Claude or OpenAI models, with your own key.
Who pays for the AI?
You use your own Claude or OpenAI key, so AI usage is billed to your account with that provider. Each platform has a cost limit per ticket, each chatbot has a daily question limit, and the Metrics page shows the AI cost per ticket.
How much does Cerberius Agents cost?
It's free on one platform, for good. The paid plans are Starter, Team and Business, monthly or yearly; the pricing page shows today's prices in your currency. AI usage is always paid directly to the AI provider through your own key.
How do we get access?
Sign up free on this site with Google, Apple or a code sent to your email, or choose a paid plan on the pricing page. Either way that sets up your organisation and its first platform, and a short guide walks you through the rest. If sign-up shows as closed, request early access and we'll set you up. Your owners and admins then invite your team.
Who on our team can do what?
There are four roles. Owners and admins set things up. Approvers review, edit and approve replies, fixes and changes. Viewers can only look. Each person has one role. Owners and admins see every platform. Approvers and viewers see only the platforms you give them.
Is there a phone app?
Yes, for Android, on Google Play. It shows what waits for you across your platforms, lets you act on it, and sends notifications. The iPhone app is on the way.
What are Development, the content writer and setup assistants?
Three more ways to use the agent, on the Team and Business plans. Development lets the people you choose ask the agent to change your product. It opens a pull request, and merges it when they say so. The content writer drafts blog articles from what your product really does. Nothing is published until a person approves it. Setup assistants let your customers set things up by chatting, inside your portal. Each change waits for the customer to press Approve.
Where do we get help?
Start with the help centre, which covers setup, daily work and common problems. For anything else, use the contact form and choose Support, or email [email protected] for sales questions.
Plans and billing
What counts as a ticket?
Each ticket the agent starts work on: a support ticket from your ticket system, an error in your logs that the log watcher turned into a ticket, or a Development request. An imported ticket counts only once you start the agent on it. Talking to the agent about a ticket it has already started on doesn't count again.
How does the free trial work?
On the pricing page, choose Try for free on Starter, Team or Business to start with 30 days free. You add a card at the start, and the plan carries on after the trial unless you cancel before it ends. Choose Select instead to start paying straight away. Paying yearly costs the same as 10 months. Moving up from the Free plan starts the paid plan straight away.
What happens when we use all our tickets?
On a paid plan, the agent carries on, and each ticket over your allowance costs US$0.50 (A$0.75). On the Free plan, the agent stops starting new tickets until the next month, or until you upgrade. Each ticket it didn't start says so.
Can we cancel?
Yes, any time, under Your account › Plan & billing. Your plan stays on until the end of the period you've paid for, and you aren't charged again.
How do we pay, and in which currency?
By card or PayPal. Prices are in Australian dollars if you sign up from Australia, and in US dollars everywhere else. The currency stays the same after you sign up. Every invoice is in Your account › Invoices as a PDF.
Still have a question?
Ask us directly, or request early access and we'll talk it through with you.