What Cerberius Agents does

Cerberius Agents is an AI support agent for the systems you look after: a product you run, your clients' systems, or your own company's IT. When a support ticket comes in, the agent looks into it the way an engineer would. It reads what you connect: your code, your Kubernetes cluster, your logs, your 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.

Each product or client is its own platform, with its own connections and tickets. The agent working on a ticket sees only that platform's systems.

How a ticket flows through#

1. 1

   A ticket arrives. Your ticket system (HaloPSA, or your own through our API) sends it to Cerberius Agents, or the log watcher opens one when it spots a new error in your logs.

2. 2

   The agent investigates. It only reads. It searches your code, and checks your cluster, logs, databases, Microsoft 365, Google Workspace and DNS records if they're connected. This usually takes a few minutes.

3. 3

   You review the draft. The ticket appears in the Queue, under Waiting for you, with the agent's findings, the evidence it found and a draft reply. You can approve it, edit it first, reject it, or ask the agent questions about it.

4. 4

   The reply goes out. When you approve, the reply is sent back to your ticket system, which passes it on to the person who asked. If the platform has no ticket system, approving simply records that you agree and marks the ticket done.

What else it can do#

• Propose a code fix. If a code change would solve the problem, the agent describes it. If you approve the fix, the agent makes the change and we open a pull request (on GitLab, a merge request) in your repository. Your team reviews it and decides whether to merge. We merge only if you ask.
• Propose a change. For example, correcting a few rows in a database, restarting a stuck service, unblocking a sign-in in Microsoft 365 or Google Workspace, or adding a DNS record at GoDaddy. You see exactly what will run, and can edit it first. Nothing runs until a person approves it. Each kind is off until your admin switches it on.
• Run checks on its own machine. If your admin gives it one, the agent gets a separate computer for each ticket, where it can look up a domain's mail records, test a mail server or check a certificate. It holds none of your keys. It's off until switched on. See Agent settings.
• Watch your logs. The log watcher checks your logs every 5 minutes. Each new, distinct error becomes a ticket, and the agent investigates it straight away.
• Answer website visitors. Chatbots answer questions on your website or customer portal from help sources you choose, such as your help pages.
• Write articles for your website. The content writer researches your market, suggests topics and drafts articles from what your product really does. Nothing is published until a person approves it. See Content writer.
• Tell your team. Telegram and Slack messages let your team know when a reply is ready, a change needs approval, an investigation failed, or an article is ready to approve.

What it never does on its own#

• It never changes anything while investigating. Investigations are read-only. The agent has no shell and no web browser, and can't run your code. It reaches the internet only through its own machine, if your admin switches that on, and that machine holds none of your keys.
• It never sends a reply without a person approving it, unless you switch on automatic replies for that platform. Even then, anything long or technical-looking is held back for a person.
• It never merges code on its own. A fix is always a pull request for your team to review. It is merged only if an approver asks.
• It never makes a change (database, cluster, Microsoft 365, Google Workspace or DNS) that a person hasn't approved, exactly as shown.
• It never sees your secrets or change logins. Approved changes are carried out by the service. For Microsoft 365 and Google Workspace, the agent gets a pass that lasts about an hour, used only by tools that read.

Where to go next#

• New here? Follow First-time setup.
• Not sure what a word means? See Key terms.
• Want the detail on safety? See What the agent can and cannot do.