You shape an answer before sending (thinking, style, instructions) and work with it afterwards: you can regenerate it, hand it to another model, export it or pick up from there in a new chat.
Thinking and styleLink to the Thinking and style section
In the model selector, “Thinking and style” sets two things for all your conversations. For a model whose thinking can’t be adjusted, the row is called “Style”.
Thinking levelLink to the Thinking level section
| Level | What it does |
|---|---|
| “Fast” | The model thinks little or not at all and answers right away. |
| “Thinking” | The model thinks before answering. |
| “Deep” | The model thinks longer on hard questions. The answer counts more toward your limit. |
The level is offered for a model that can think, as well as in “Auto” mode and in a comparison, where it applies to each model that can think. “Default” lets each model think as it usually does.
Answer styleLink to the Answer style section
| Style | What it does |
|---|---|
| “Normal” | Balanced answers. |
| “Concise” | Gets straight to the point. |
| “Explanatory” | Explains step by step, with examples. |
| “Formal” | Formal, professional tone. |
To create your own style:
- In “Thinking and style”, click “Create style”.
- Paste a text whose tone you like (an email, an article, a post), then click “Analyze text”. Moon AI turns it into a description that you review and edit. The text goes through Moon Blur and is not saved.
- Or click “Write the description myself”.
- Name the style, then click “Save”.
You can keep up to 5 personal styles. Deleting a style doesn’t change answers already written.
Conversation settingsLink to the Conversation settings section
“Conversation settings”, in the ⋯ menu at the top of the conversation, opens the “Chat controls” panel:
| Setting | What it does |
|---|---|
| “Instructions for this conversation” | How Moon AI should answer you here: tone, role, format. “Start from an example” suggests instructions for you to complete. |
| “Creativity” | “Precise”, “Balanced” or “Creative”: more predictable or more varied answers, for this conversation only. |
| “Response style” and “Thinking level” | The same settings as in the selector, for all your conversations. |
| “Files” | The files attached to the conversation. Remove one so that Moon AI stops using it. |
For instructions that apply to all your conversations, see custom instructions.
Stop, continue, retryLink to the Stop, continue, retry section
- While the answer is being written, “Stop” replaces the send button. The start of the answer stays on screen. To pick it up again: “More actions” › “Continue response”.
- When an answer fails, “Retry” runs it again. If some steps had already succeeded (a search, an action in a tool), the button becomes “Resume” and continues without redoing those steps.
An answer cut off while thinkingLink to the An answer cut off while thinking section
A model that thinks a lot can use the whole length allowed for one step of the answer on thinking, with no time left to write. Moon then runs that step again once, with less thinking, without asking you. The retry counts toward your usage limit as one step of the answer.
To avoid hitting the wall again, the following messages in the same conversation also go out with less thinking for two hours.
If the answer is still cut off, a line at the end says so: “Answer cut by the length limit: the model spent all the room of this step thinking and had no time to write. Send “continue” to resume.” Then send “continue” so the model picks up its answer.
Edit your messageLink to the Edit your message section
- Below your message, click “Edit”.
- Fix the text.
- “Send” asks for a new answer; “Save” keeps the fix without running it again.
“Copy” copies your message.
Versions and branchesLink to the Versions and branches section
Each edit of a message and each regeneration create a new version, without erasing the previous one. Below the message, the “Previous message” and “Next message” arrows and the counter (“2/3”) move from one version to another. Each version keeps the part of the exchange that goes with it: the conversation then forms several branches.
The conversation continues from the version shown. To carry on with a branch in a separate conversation, use “More actions” › “New chat from here”. The new chat picks up the exchange up to that answer, in the same project, with the title followed by “(continued)”.
Actions below an answerLink to the Actions below an answer section
Below each finished answer:
| Button | What it does |
|---|---|
| “Copy” | Copies the answer. The “Copy formatted text” setting (“Settings” › “General” › “Answers”) keeps the formatting. |
| “Good response” / “Bad response” | Rates the answer. A card then lets you say why (“Why?”). |
| “Regenerate” | Opens the answer variants (below). |
| “Read aloud” | Reads the answer out loud. A second click stops it (see Voice). |
| “Add a tag” | Marks the answer “Important”, “To review” or “Done”. Depending on your plan. |
| “More actions” | Opens the ⋯ menu (below). |
RegenerateLink to the Regenerate section
“Regenerate” offers:
- “Try again”: another answer to the same question;
- “Add details”, “More concise”, “Simpler”, “More professional”;
- “Try another model”: you choose the model that answers this time;
- “Try again with web search”;
- the “Suggest a change” field, to say what should change, then “Send”.
Each variant becomes a new version of the answer (see Versions and branches).
The “More actions” menuLink to the The “More actions” menu section
Depending on the answer, the ⋯ menu offers:
| Item | What it does |
|---|---|
| “Edit” | Fixes the text of the answer. |
| “Open in editor” | Opens an editable copy next to the conversation (see The editor). The answer doesn’t change. |
| “Audio overview” | Two voices explain this answer (see Audio overview). |
| “Continue response” | Picks up an answer that was stopped or cut off. |
| “New chat from here” | Opens a conversation that picks up the exchange up to this answer. |
| “Export” | “Word document (.docx)”, “PDF document (.pdf)”, “Gmail draft” (opens a new message in Gmail) or “Google Docs” (copies the answer so you can paste it into a new document). |
| “Verify answer” | Runs one web search per claim (see Check an answer). |
| “Report this answer” | Reports the answer to the Moon AI team (see Report an answer or content). |
| “Delete” | Deletes this version of the answer, when there are several. |
On a phoneLink to the On a phone section
A long press on a message opens the “Message actions” sheet: “Copy”, “Select text” (then “Copy selection”), “Edit”, “Regenerate”, “Read aloud”, “Export”, “Open in editor” and “Delete”.
On a passageLink to the On a passage section
Select a passage of an answer: two buttons, “Ask” and “Explain”, ask a question about that passage. The “Floating quick actions” setting (“Settings” › “General” › “Answers”) turns them on or off.
Report an answer or contentLink to the Report an answer or content section
If an answer, an image, a video or a sound created by the AI seems offensive, dangerous or false to you, report it to the Moon AI team without leaving the conversation. It works the same way on a computer, on a phone and in the apps.
| Content | Where to report it |
|---|---|
| An answer | Below the answer, “More actions” › “Report this answer”. |
| An image created by Moon AI | Click the image to open it full size: the “Report this image” flag is at the top of the viewer, next to “Download image” and “Copy image”. |
| A video | The “Report this video” flag, below the player, next to “Download video”. |
| Music or an audio overview | The “Report this audio” flag, below the player, next to “Download audio”. |
The window that opens says: “Tell us what is wrong with this content. The Moon AI team reviews every report.”
- Under “Reason”, choose one: “Offensive or hateful content”, “Sexual content”, “Violence or dangerous content”, “False or misleading information” or “Other”. A reason is required.
- If you like, add details in “Comment (optional)”, in 1,000 characters at most.
- Click “Send report”.
“Thank you. Your report was sent to the Moon AI team.” confirms it was sent. Reporting the same content again doesn’t create a second report. The number of reports is limited per hour.
What a report containsLink to the What a report contains section
For moderation, the report keeps a copy of the reported content: the text of the answer (its first 4,000 characters), or the link to the image, video or sound. It also includes your account, the conversation, the model that answered, the reason and your comment.
In a confidential chat, reporting is offered too, and the window says so before sending: a copy of the reported content goes to the Moon AI team with the report. The rest of the conversation is neither sent nor kept.
The report stays on record if you delete the conversation afterwards. It is included in your data export and erased with your account.
When reporting isn’t offeredLink to the When reporting isn’t offered section
- A shared conversation, opened read-only, doesn’t offer reporting.
- While an answer is being written: the option appears once the answer is finished.
Code and diagramsLink to the Code and diagrams section
A code block shows its language in the header, with “Copy code”, and “Collapse” or “Expand”. A long block is collapsed, with “Show all N lines”. The “Always collapse code blocks” setting (“Settings” › “General” › “Answers”) collapses every block. Ctrl/⌘Shift; copies the conversation’s last code block.
A diagram or chart that the AI describes in code is shown drawn, with zoom. In a conversation that contains content from outside, you need to click before it is shown.
Questions from MoonLink to the Questions from Moon section
When your request leaves a choice open, Moon can ask you a question with answers to click (a “Question for you” card):
- Choose one answer, several, or rank them by drag and drop, depending on the question.
- “Other” lets you write your own answer.
- Click “Send” (“Send ranking” for a ranking).
You can also answer in the message box.
Visuals in the answerLink to the Visuals in the answer section
An answer can contain an “Interactive visual” (chart, diagram, small mockup) drawn in place. Below the visual:
| Button | What it does |
|---|---|
| “View code” / “View visual” | Switches between the rendering and the code. |
| “Copy as image” | Copies the visual as an image. |
| “Save as artifact” | Files it in the artifacts panel so you can find it and edit it. |
| “Open in full screen” | Enlarges the visual. |
After a web page, a file or a tool result in the conversation, a visual doesn’t run on its own: “Not run automatically” appears with a “Run” button. Check its code before running it.
A suggestion toward Code, Moon Desktop, a routine or a deep researchLink to the A suggestion toward Code, Moon Desktop, a routine or a deep research section
When your request would be better handled elsewhere, a small card can appear below the answer. The answer is always given in Chat first: the card never blocks anything.
| Card | When | Button |
|---|---|---|
| “This would be simpler in Code” | You ask to work in a project or folder on your computer: fix a project, run its tests, a Git command, organize files. | “Open Code” if Moon CLI is connected to your account, otherwise “Install Moon Desktop” and “Explore Code”. In Moon Desktop, “Open in Code” lets you choose the folder. |
| “This would be better with Moon Desktop” | On the website, you ask for something only the desktop app can do: a screenshot, a file on your computer that you didn’t attach, a quick window. | “Get Moon Desktop”. |
| “Moon can do this for you regularly” | You ask for a task that repeats (“every morning”, “every Monday”) or to be told when something happens. | “Create a routine”. |
| “Go deeper with a sourced report” | Your question calls for real research: the state of a market, its main players, a comparison of offers or tools, recent figures, sources. Only if you still have a deep research left this month. | “Start deep research”. See Deep research. |
If your plan doesn’t include Code, the card says so (“Code is included in a higher plan.”) and offers “View plans”, never a price. On a phone, only the routine and deep research cards appear.
- Low-key: one card per message at most, each card only once per conversation, and two cards a day at most. Nothing in a temporary conversation, or in a shared conversation opened read-only.
- Nothing sent: the card is chosen on your device, based on the text of your message; no text from the conversation is sent to choose it.
- Your call: the cross (“Hide this suggestion”) closes the card. “Do not suggest this again” stops this kind of suggestion for your account, on all your devices.
See Code in Moon Desktop, Code on the web, Routines and Deep research.
Tool detailsLink to the Tool details section
When the AI uses tools (web search, reading a page, an isolated computer command, a connector), a collapsed block above the text sums up the steps in one line. Clicking it opens it: each step, what was asked and what came back.
Creations (images, files, documents) appear below the text of the answer.
Long conversationsLink to the Long conversations section
When a conversation goes beyond what the model can read, Moon AI sends it a summary of the beginning. A line on the answer says so: “Start of the conversation summarized to fit the model’s memory”. Older messages stay fully visible; only what the model receives changes. For a very long document, the model reads the most useful passages and says so.
Display while the answer is writtenLink to the Display while the answer is written section
“Soft fade-in for text as it is written” (“Settings” › “General” › “Answers”) makes the last line appear gradually. “Follow-up auto-generation” (“General” › “Conversations”) suggests questions at the end of an answer.