Contents

FAQ

FAQ

Frequently asked questions (common questions, questions and answers) about AnyLLM — short answers with a link to the page that explains the topic in full. The first part is for everyone who uses the chat, the second for the administrators who set it up. Error messages and their remedies are on Troubleshooting.

FAQ for users

Where do I find AnyLLM? In Jira and Confluence under Apps → AnyLLM Chat, on Jira issues (panel, sidebar panel, ••• action, issue navigator), on Confluence pages (macro, assistant below the title, chat bar, quick actions) and in Jira dashboards (gadget) — see Where AnyLLM appears.

Does the AI see confidential content? Only what you may see yourself: every lookup runs with your own Jira and Confluence permissions. What you type, your memory notes, the page or issue context and the site knowledge do go to the AI service your organisation has set up — see What goes where.

Can anyone read my conversations? No. Conversations, memory and your own skills are visible to you alone — the administration cannot open them either. Something becomes visible only when you share it yourself: by archiving a chat to Confluence, sharing an agent or via a dashboard gadget — see Who can see what.

How do I prevent something from being saved? Switch on Temporary chat (not saved) in the ＋ menu; the chip Temporary ✕ shows that it is active and ends it with a click — see Temporary chat.

How do I keep a chat permanently? Pin it so that it is not dropped when your limit of 50 conversations is reached, archive it to Confluence (a plain-text copy in your personal space) or export it as Markdown — see Managing Chat History.

Why does the AI answer in a different language than expected? It answers in the language of your latest message. Write in the language you want, or ask for it explicitly — see Chatting with the AI.

Why are some messages in English although the interface is German? Messages from the server are always English; only the texts of the interface follow the interface language. Troubleshooting quotes them verbatim.

Can the AI create or change issues and pages on its own? No. It only produces a suggestion card (Create Jira issue, Update Jira issue, Add Jira comment, Create Confluence page); nothing happens until you click ✓ Execute — see Letting the AI Search, Read and Create.

Why is my agent not in Confluence although I created it in Jira? It should be: your own agents, skills, memory and history belong to your account and are identical in Jira and Confluence. If something is missing, reload the page or ask your administration whether Confluence is connected to the app — do not create the agent again — see Creating Your Own Agents and Skills.

Can I share my agents with colleagues? Yes, with the checkbox Make available to all users of this site: the agent then appears under Shared by colleagues for everyone on the site, in both products; others can use it but not edit it, and your instructions stay hidden. Skills cannot be shared — see Sharing an agent.

Why do I see only one model, or no agent selector at all? The model selector appears only when more than one model is enabled, the agent selector only when at least one agent is available to you — see Choosing agent and model.

Answers are too short or too long — can I control that? Yes: ＋ → ⚙ Model parameters … lets you set Temperature and Max response tokens; the choice is remembered for you — see Model parameters.

What does ⤢ Fullscreen do? It opens the current chat as a fullscreen overlay titled AnyLLM on top of the issue or page you are on — it looks like the app page, and you close it with the overlay's own ✕. Only if the overlay cannot open are you taken to the app page instead — see Fullscreen overlay and Fullscreen overlay in Confluence.

What is the standalone full page? The chat on its own browser page without the Jira or Confluence frame. There is no menu entry for it — the only way in is Open as full page at the bottom of the app-page sidebar — see Standalone full page.

Why does a quick-action result appear in my history? Every result of the Confluence quick actions (on selected text and on the whole page) is saved as a new conversation in your history so that you can carry on with it; Continue in chat → opens the app page, where you find it in the sidebar — see Quick actions on selected text.

Which conversation does "Continue conversation" in the chat bar open? The one you last worked in — the conversation with the most recent activity, regardless of pinning — see Chat bar at the top of pages.

Does the AI remember everything in a long chat? No — only the most recent 16 messages (about 32,000 characters) are sent to the AI, although the history keeps everything. Repeat important facts or start a new chat with a short summary — see What the AI remembers within a conversation.

What happens when I click Stop? Only the display stops; the request continues on the server, and the complete answer may still land in the saved conversation. Delete the conversation afterwards if you do not want that — see Stopping an answer.

FAQ for administrators

Do I have to set up Jira and Confluence separately? No. AnyLLM is one installation with one configuration for both products — connections, agents, templates, module toggles and general settings apply everywhere. The one exception is the endpoint-approval consent in Atlassian's confirmation dialog, which counts per product: grant it once in Jira and once in Confluence — see Administration & Configuration and How approvals work per product.

Which AI services can I connect? Any OpenAI-compatible service (LiteLLM, OpenAI, OpenRouter, self-hosted servers such as vLLM, Ollama or LM Studio), Azure OpenAI and Anthropic — up to 5 connections per installation, each with its own https address, API key and model list. Every address needs an endpoint approval first — see Setting Up LLM Connections.

Where are the API keys stored? Encrypted in Atlassian's secret store of this installation, one per connection (plus one for the search service). They are readable only by the app backend and can be replaced or deleted, never read out — see Data storage.

Can I see who consumes how much? No. Usage shows site-wide totals per model for the current and the previous month; nothing in the interface attributes requests to a person (the app keeps a per-person monthly counter in its storage that nobody can view and that is deleted with the account) — see Reading the usage statistics.

Can I view or export chat content? No. Conversations are accessible only to the person who had them; there is no central export — see Who can see what.

What happens on uninstallation? Atlassian deletes the whole app storage of the installation — both products, all data, including the API keys. Issues, comments and pages created via suggestion cards and archived chats remain, because they are normal Jira and Confluence content — see Three deletion paths.

Why do users still see disabled surfaces in the menu? Menu entries cannot be hidden per installation. The entry stays visible and opens the notice "This feature has been disabled by your administrator."; the server rejects requests from disabled surfaces anyway. Only the chat bar disappears completely — see What users notice when a module is off.

Why does the chat bar not disappear — or reappear — immediately? Its status is cached in each browser for ten minutes, in both directions: after you switch it off it may stay visible, and after you switch it on or complete the setup it may stay hidden for up to ten minutes — see Chat bar at the top of pages.

Can I prevent content from going to the AI service? You decide which service is connected and what the AI may read: with AI access to Jira & Confluence set to Off, no issue or page content leaves via the app. Users' chat inputs, their memory notes and the site knowledge still go to the service, and web-search queries go to the search service you enabled — see AI access to Jira & Confluence.

How do I prove to data protection or auditing where data flows? The list under Endpoint approvals is complete: without an entry there the app reaches no external address, and no address is pre-approved. For operations and privacy documentation beyond this manual, ask the vendor maplee via your administration — see Checking approvals.

How do I back up the configuration? With ↓ Export as JSON on Import/Export: connections (without keys), agents, templates and general settings. API keys, conversations and memory cannot be backed up centrally — see Exporting the configuration.

How do I change an API key without downtime? Enter the new key in the connection's API key field and click Save connections — the new key applies immediately; an empty field keeps the old one — see Changing or deleting an API key.

How do I move an endpoint? Change the connection's Endpoint, click Save connections and confirm the Atlassian dialog for the new address (in both products). After a successful save the app automatically revokes approvals that no saved connection or search service uses any more — the old address needs no manual revocation — see Automatic revocation after saving connections.

Why does the chat work in Jira but not in Confluence? Almost always because the endpoint approval is missing in Confluence (the consent counts per product) or because Confluence is not connected to the app — see Common misconfigurations and recovery.

Why do the usage statistics stay empty? The AI service must return token usage in its streaming responses; if it does not, the app has nothing to count. Ask the provider whether its streaming API returns usage — see Reading the usage statistics.

What happens to the data of a person who leaves the company? Once the Atlassian account is closed, a weekly automatic run deletes everything the app stored for that person — conversations, memory, preferences, own agents and skills, usage counters and the agents they shared. There is nothing to set up, and the run cannot be switched off — see Three deletion paths.

What does an inactive subscription block? Only new chat answers and ✓ Execute on suggestion cards. History, memory, own agents and skills and all settings pages including Test connection keep working — see Licence and subscription.

Documentation baseline: app version 0.2.0 · 2026-08-30