Chatting with the AI

This page covers everyday chatting with the AI (asking questions, prompts, answers): sending a question, what the states while the answer is generated mean, choosing an agent and a model, regenerating or editing, stopping an answer, exporting a conversation, the blue chips, what AnyLLM remembers for you and what the AI remembers within a conversation.

What this does

You type a question, the AI answers while the text is generated, and the conversation is saved automatically in your history. Where the surface knows an issue or a page, that content can be included; where the administration allows it, the AI can also look things up in Jira and Confluence — see Letting the AI Search, Read and Create.

Who can do this

All users. Every request runs with your own Jira and Confluence permissions.

Before you start

• The app is set up, the surface you are using is enabled and — for a paid installation — the subscription is active (see Getting Started).
• You have opened AnyLLM on any surface, e.g. Apps → AnyLLM Chat.

Step by step: asking a question

1. Click into the field Message AnyLLM ….
2. Type your question. For a line break press Shift+Enter.
3. Press Enter or click the send button (arrow icon, tooltip Send (Enter)).
4. Wait for the answer. While it is generated, the field is locked and the send button becomes the stop button (tooltip Stop).

Expected result — the answer appears word by word. Below the finished answer you see the name of the model that produced it. The conversation appears under History and, after the first answer, receives a short title generated by the AI (at most six words); if that fails, the first question is used as the title.

What the states mean

Notes

• Answers may contain Markdown: headings, lists, tables, code blocks and links. If an answer contains a diagram in Mermaid notation (for example because you asked for one), it is rendered as a graphic.
• Images in answers are not loaded automatically. Instead of the image you see the link view external image; a click opens it in a new tab. This prevents an image from reporting data to the outside unasked.
• The automatic save does not apply to a temporary chat — see Temporary chat.

Choosing agent and model

In the bar below the input field, next to ＋, there are up to two selectors:

• The agent selector (screen-reader label Choose agent) — only appears if at least one agent is available to you (predefined by the administration, shared by colleagues or your own). No agent means: general chat. If the agent has a fixed model, the model selection switches with it. More: Creating Your Own Agents and Skills.
• The model selector (screen-reader label Choose model) — only appears if more than one model is enabled. The models are grouped by connection.

Both selections are remembered for you (see Remembered choices). If you reopen a saved conversation, the model and the agent of that conversation are restored, provided both still exist.

Regenerating an answer or editing a message

• ↻ Regenerate answer appears below the last answer as soon as the AI has finished. The last answer — including any open suggestion cards — is discarded and the question is sent again. Each regeneration counts as a new request.
• Edit & resend — hover over one of your own messages and click the pencil icon. The text returns to the input field, and the conversation is cut off from that point: everything after it is discarded. Adjust the text and send again.

Stopping an answer

While the AI is writing, the send button turns into the stop button (tooltip Stop). A click hides the running answer, and you can write again immediately.

⚠️ Warning: Stop only ends the display. The request continues on the server, and the complete answer can still land in the saved history. If you do not want that, delete the conversation afterwards — see Managing Chat History.

Exporting a conversation

• On the app page, in the fullscreen overlay and on the standalone full page: click ＋ and then ↓ Export conversation (Markdown) (shown once the conversation has messages).
• In dialogs and panels (issue panel, issue sidebar panel, issue action, page assistant, macro, issue navigator): click ↓ Export in the header.

Expected result — a Markdown file named llm-chat-YYYY-MM-DD.md is downloaded. It contains your questions, the answers and the model names used; suggestion cards are not exported. The export runs entirely in your browser; nothing is transmitted.

The blue chips

In the bar below the input field, to the left of the send button, small blue chips show what is currently active:

Remembered choices

The model, the agent, the setting Include page context and your model parameters are saved for you and restored the next time you open the chat. Because AnyLLM is one installation for both products, the same choices apply in Jira, in Confluence, in the fullscreen overlay and on the standalone full page.

What the AI remembers within a conversation

• Only the most recent 16 messages of a conversation (at most about 32,000 characters) are sent to the AI. In long chats the AI no longer remembers earlier turns even though they stay in the history. If an early detail matters, repeat it.
• Your memory notes and the administration's site knowledge are sent with every request; page or issue context is re-read with your permissions for every message.
• Per answer the AI may look things up in at most 5 rounds (Jira/Confluence searches, reads, web search) before it must answer. Each round may take up to 120 seconds; the whole answer at most 5 minutes; each lookup result is cut at 8,000 characters. Split very long multi-step tasks into smaller questions.
• If the AI only produced a suggestion card and no text, the answer reads I proposed an action — please confirm or dismiss it above.

Common problems

Related topics

• Getting Started
• Prompts, Skills, Memory and Model Parameters (＋ Menu)
• Creating Your Own Agents and Skills
• Letting the AI Search, Read and Create in Jira and Confluence
• Managing Chat History: Pin, Fork, Archive, Delete
• Permissions, Roles and Data Visibility
• Troubleshooting
• FAQ