AI chat workflows

This page focuses on practical ways to get useful results from AI in everyday map work.

Benefits and risks

Benefits:

workflow patterns can make repeated map tasks faster and more consistent,
profiles and prompts reduce repeated setup work.

Risks:

AI suggestions can still be wrong,
hidden prompt runs reduce visibility,
tool-call summaries help review what happened, but they do not prove that the result is correct or safe.

Use profiles and prompts for repeatable work

Use Edit profiles... when you want reusable chat behavior across many requests. Select the profile in the AI profile control before sending your request.
Use prompts when you want saved actions you can launch directly from menus, for example to rewrite the selected node, summarize a branch, draft a decision, or run a hidden background request.
Unlike normal chat, prompts do not use the current assistant profile, and hidden prompts can run without replacing the currently visible chat.

For details, see AI prompts and profiles.

Reuse prompts inside normal chat

To reuse a saved prompt without launching it from a menu, start the chat input with / and the prompt name.

Example:

/Summarize branch for release notes

If the leading /Summarize branch text matches a saved prompt, Freeplane keeps that prompt reference visible in the chat but sends the saved prompt text to the model, followed by the remaining text. If no saved prompt matches, the message is sent as ordinary chat text.

Prompt references use only the saved prompt text. They do not apply the prompt's saved visibility, model, temperature, thinking-effort, or tool settings.

Choose model parameters for interactive chat

The AI panel top bar contains the model selector and the thinking-effort selector. The panel menu contains the Temperature submenu.

Model selection chooses the provider/model used for ordinary chat.
Thinking effort uses Freeplane's provider-independent levels: Max, Extra high, High, Medium, Low, Minimal, and None.
Temperature can be Model default, one of the numeric presets, or a custom numeric value.

Providers and models do not all support the same thinking controls. When needed, Freeplane maps the selected thinking effort to the closest available provider behavior, or the provider may ignore it.

Review chat instructions

The AI chat options menu has two independent instruction controls:

Show instruction history shows the committed system and profile instruction blocks stored with the chat. This is useful when you want to inspect which instructions were active for earlier visible chat requests.
Preview next request instructions shows a non-persistent preview of the system and profile instructions that would be used by the next visible request.

The preview appears below the AI profile selector and above the input area. It updates when relevant request inputs change, for example the selected profile, tool availability, system instruction state, or a leading slash prompt reference in the input.

The preview is not part of the chat transcript. It is not copied with chat history, saved as a committed chat message, counted as a chat turn, or sent as a separate user message. It is a display of the next request's instruction state.

If the selected profile differs from the profile currently active in the chat, the preview applies that pending profile change first. The preview then shows the composed system instruction followed by the effective profile instruction. If the draft starts with a resolved slash prompt reference, the preview also includes the saved prompt text that would be substituted into the request.