Secrets and your data
How keys and logins are protected, what Cerberius Agents stores, how long it keeps it, and who can see it.
Keys, tokens and logins#
Everything secret you give Cerberius Agents is stored encrypted. That includes:
- Claude, OpenAI, GitHub and GitLab keys;
- Kubernetes logins (kubeconfigs) and the Loki login;
- database passwords, for both read logins and change logins;
- Telegram bot tokens and Slack webhook addresses;
- webhook signing secrets.
How they're protected:
- Encrypted at rest. Each saved value gets its own encryption key, which is itself encrypted. Each value is tied to your organisation and to the item it belongs to, so it can't be copied to another place and used there.
- Never shown again. Once saved, a secret can't be read back, in the app or through the API. You only see its last four characters. To change it, you replace it.
- Used only where it's needed. Keys in your library can only be used by your own organisation's platforms. Change logins and write access to your code are used only by the service when a person approves a change. The agent never receives them.
With your own runner
When the agent runs on your own runner, the keys into your own systems (the agent's AI key, repository keys, cluster, Loki and database logins, Microsoft 365, Google and GoDaddy) are kept by you, and we never hold a copy we can read. Only the keys our service uses itself stay with us, encrypted. See Where your keys are kept.
Give each key the least access
Use read-only keys and logins wherever you can. Add write access only where you've switched on fixes or changes. See Keys.
Secrets hidden from what's stored#
Before tool output, error messages and logs are stored, Cerberius Agents hides values that look like secrets. That covers:
- private keys;
- passwords and tokens in "key = value" settings and in authorisation headers;
- well-known token formats, such as Claude, OpenAI, GitHub, Slack, Telegram, AWS, Google and Stripe keys, and sign-in tokens;
- card numbers.
They're replaced with [REDACTED]. Cluster, log and database results are cleaned the same way before the agent sees them.
Ordinary details, such as names, email addresses and booking references, are kept, because investigations need them.
What's stored#
For each platform, Cerberius Agents keeps:
| What | Why |
|---|---|
| Tickets, follow-ups and attachments | So the agent and your team can work on them. |
| Drafts and replies, with who approved, edited or rejected them | Your record of what was sent and why. |
| The agent's findings and the steps it took | So you can see how it reached its answer, in each ticket's What happened. |
| The agent's working record for each ticket | So it can pick up where it left off when you ask it something or a follow-up arrives. |
| Changes to approve and their results | A record of what ran and what happened. |
| Chatbot conversations | So you can review how your chatbots answer. |
| A log of setting changes | Who added a key, removed someone from the team, deleted a platform, and so on. |
Your code is downloaded fresh for each run and isn't kept afterwards.
How long it's kept#
Ticket data is kept until you delete the platform it belongs to. Deleting a platform removes all of its tickets, history, settings and chatbots for good. See Platforms.
There's no automatic clean-up of old tickets yet.
Who can see it#
- In the app, only your organisation's team can see your tickets, settings, keys and team. What they can do depends on their role. See Members and roles.
- Each organisation is kept separate. Nobody from another organisation can see or use anything of yours.
- Within your organisation, each platform's tickets and settings are separate. The team, keys, rules and notification channels are shared by all your platforms. Approvers and viewers only see the platforms switched on for them.
- Notifications carry very little. Telegram and Slack messages show only the platform, the ticket subject, who sent it and a link. Never findings, code or logs.
The AI provider#
To investigate, the agent sends the ticket and what it reads (code, logs, query results) to Claude, using the Claude key you chose. Chatbots send visitors' questions and your help content to Claude or OpenAI, using the key you chose. That use falls under your own account with that provider.