Key terms
Plain definitions of the words used across Cerberius Agents.
These are the words you'll see across the app and in this help.
Who's who#
| Term | What it means |
|---|---|
| Organisation | Your company in Cerberius Agents. It holds your team, keys, rules, notification channels and plan, and one or more platforms. |
| Platform | One system you support, such as your booking app, or one client you look after. Each platform has its own tickets, connections, settings and chatbots. |
| Member | A person from your organisation who can sign in. Each member has one role, which applies to every platform. |
| Owner, Admin, Approver, Viewer | The four roles. Owners and admins set things up, approvers review and approve, viewers only look. See Members and roles. |
| Approver | Anyone who can approve replies: owners, admins and approvers. |
| Requester | Whoever opened the support ticket: usually a user of your product, or a client's staff member. Requesters never sign in to Cerberius Agents. They only see the replies you approve. |
Tickets and replies#
| Term | What it means |
|---|---|
| Ticket | One support request. It comes from your ticket system (HaloPSA or your own), or from the log watcher when it finds a new error. |
| Ticket number | Our own number for a ticket: #1, #2 and so on, counted separately on each platform. Your ticket system's own number shows next to it. |
| Investigation | The agent's read-only look into a ticket: your code, and your cluster, logs and databases if connected. |
| Run | One working session of the agent on a ticket. The first investigation is a run. So is each later step, such as answering a follow-up, a question from you, or making an approved fix. What each run found or said shows in the ticket's What happened. |
| Draft | What the agent prepares for you to review. Usually a reply to the requester with the agent's findings. It waits in the Queue until someone approves or rejects it. |
| Reply | The message that goes to the requester once a draft is approved. |
| Queue | What waits for you on a platform: drafts to approve, tickets that need a person, and investigations that failed. Below them, tickets the agent is still working on. |
Fixes and changes#
| Term | What it means |
|---|---|
| Fix | A code change the agent proposes. If you approve it, the agent edits one repository and we open a pull request (GitLab: merge request). Your team decides whether to merge it. An approver merges it with Merge on the ticket, or by asking the agent. |
| Change to approve | A change the agent proposes to a database, a cluster, Microsoft 365, Google Workspace or DNS records at GoDaddy or Cloudflare, such as updating some rows, restarting a service or unblocking a sign-in. It shows exactly what will run, and runs only after a person approves it. It's never sent to the requester. |
| Change login | A separate login, with permission to change things, that the service uses only for approved changes. The agent never receives it. |
Setup#
| Term | What it means |
|---|---|
| Key library | Your organisation's saved keys, under Organisation › Keys: Claude, OpenAI, GitHub and GitLab. Each platform, repository and chatbot picks the key it uses. |
| Rules | Instructions every agent and chatbot in your organisation follows, after our own safety rules. |
| Ticket system | Your help desk, such as HaloPSA, or your own support tool, connected so tickets arrive automatically and approved replies go back to it. |
| Log watcher | A setting that checks a platform's logs every 5 minutes and opens a ticket for each new error it finds. |
| Chatbot | A chat window for your website or customer portal that answers visitors from help sources you choose. |
| Help source | A website or Markdown file a chatbot answers from. |
| Notification channel | A Telegram chat or Slack channel that hears about new work. |
| Automatic replies | An optional platform setting that lets some replies go out without approval. Off by default. |