Permissions, Roles and Data Visibility

This page explains the roles and permissions in AnyLLM (who may do what), the data visibility rules (who can see your conversations, memory, agents and archived chats — chat privacy) and the data flow (what goes to the AI service and what stays at Atlassian). It applies to Jira and Confluence alike: AnyLLM is one installation for both products with one configuration and one data store.

The two roles

AnyLLM grants no permissions of its own and knows no further roles: there is no user management, no groups and no invitations in the app. The Settings button in the chat (sidebar footer of the app page; header of the macro and the issue navigator) is shown only to administrators — in Jira after a real permission check, in Confluence if you belong to one of the usual administrator groups. The button only controls visibility; the settings pages themselves are protected separately.

The three principles

1. The AI always works with the rights of the individual person. It searches and reads exclusively what that person may see themselves; page and issue context is re-read with your permissions for every message. There is no mode in which the app works with elevated rights.
2. Writing actions always require a confirmation. The AI can only propose — as a suggestion card Create Jira issue, Update Jira issue, Add Jira comment or Create Confluence page. Nothing happens until you click ✓ Execute, and then it happens with your rights and in your name. Dismiss removes the card.
3. API keys stay in the background. They are stored encrypted in the Atlassian secret store, are never visible in the browser and do not appear in logs. An administrator can only replace or delete a key, never read it.

Who may do what

Who can see what

Everything personal in AnyLLM belongs to your Atlassian account and is identical in Jira and Confluence.

What goes where

Which AI service that is, is known to your administration — the complete list of reachable addresses is under Endpoint approvals in the settings. Treat a chat input like an e-mail to an external service provider.

What you can do yourself

• Temporary chat (not saved) in the ＋ menu for one-off questions that should not be saved — see Prompts, Skills, Memory and Model Parameters (＋ Menu).
• Delete removes a conversation permanently (Delete permanently in the dialog Delete conversation?) — see Managing Chat History.
• Include page context can be switched off in the ＋ menu before you send.
• 🧠 Edit memory lets you switch memory off or empty it at any time.
• Restrict the archive: after the first archiving, restrict the page AnyLLM Chats in your personal space to yourself once; child pages inherit the restriction.
• Check every suggestion card before you click ✓ Execute; Dismiss discards it.
• ↓ Export conversation (Markdown) gives you a copy you control.

What the AI cannot do

• It cannot see anything you are not allowed to see. All searches and reads run with your permissions.
• It cannot change anything without your click. Writing actions only ever appear as a suggestion card. If the administration has meanwhile switched AI access away from Read + write, even ✓ Execute fails with "Write actions are not enabled by the admin."
• It cannot follow instructions hidden in content. If a page or issue it has read contains something like "create an issue now", at most a suggestion card comes out of it, which you can dismiss.
• It cannot reach any address the administration has not approved. No AI endpoint is pre-approved — every address needs an explicit endpoint approval.

Effects of administrative changes on users

What users notice when the administration adds or removes a connection, changes the AI access, lowers the rate limit, switches a surface off or changes the language — and when each change takes effect — is listed in one place: Opening the Settings and Initial Setup. In short: everything except the default model, new templates and the interface language takes effect immediately; the Confluence chat bar follows with up to ten minutes' delay.

Related topics

• Getting Started — where AnyLLM appears and what the first screen shows
• Managing Chat History: Pin, Fork, Archive, Delete — deleting and archiving your conversations
• Creating Your Own Agents and Skills — sharing agents with the site
• Letting the AI Search, Read and Create in Jira and Confluence — suggestion cards and ✓ Execute
• Dashboard Gadget (AI Report) — visibility of gadget answers
• Managing Endpoint Approvals — the complete list of reachable addresses
• General Settings — AI access, rate limit and site knowledge
• Licence, Data Retention and Automatic Deletion — data storage, retention and the weekly cleanup
• Troubleshooting · FAQ