October 31, 2026Link to the October 31, 2026 section
Organization MCP servers: each member with their own accountLink to the Organization MCP servers: each member with their own account section
In a managed organization, an MCP server shared by the organization (Notion, Linear…) connects with each member’s own account: everyone acts under their own name in that service. These servers appear on the “Connectors” page, in the “From your organization” section, with “Connect”, “Reconnect”, or “Provided by your organization” when the organization provides a shared token. An action approved later always runs with the account of the person who asked for it.
October 30, 2026Link to the October 30, 2026 section
Managed organizations, and passkey sign-inLink to the Managed organizations, and passkey sign-in section
A company, a school or a public administration can now have its organization managed in Moon AI, under contract. The owner and the admins open a console with “Administer …” from the workspace menu: members and groups, licenses, rules, the organization’s models and MCP servers, single sign-on (SSO) and provisioning (SCIM). A workspace under contract says so on its “Billing” page, with no price and no invoice to pay online.
For all accounts:
- passwordless sign-in with a passkey (fingerprint, face or device lock): “Add a passkey” in “Settings” › “Security”, then “Sign in with a passkey” (on the night of October 30 to 31);
- “Settings” › “Account” shows whether your email address is verified and lets you verify it right there;
- “Manage workspace” becomes “Your workspace”, and a “Usage” tab in settings breaks down your usage by day and by model.
See Managed organizations, SSO and Security.
October 29, 2026Link to the October 29, 2026 section
A new interfaceLink to the A new interface section
The whole interface has been redesigned: fonts, colors, sidebar, home screen, conversation thread, model picker, menus and every page (Agents, Library, Notes, Projects, Settings, Subscription, sign-in, error pages).
- Lune, the mascot, keeps you company on empty screens, errors and waits. It never appears next to another provider’s model: models keep their real logos.
- A browser that has never picked a theme opens in the “Night” theme.
- Moon AI emails take on the same colors.
Your conversations, files and settings stay the same.
October 28, 2026Link to the October 28, 2026 section
An avatar for every agentLink to the An avatar for every agent section
In “Create an agent” and “Edit the agent”, choose a profile (the mascot in twenty outfits) set on a planet (twenty backgrounds), or use “Random”. Moon’s agents each have their own, and your existing agents got an avatar on the planet of their former color.
See Agents.
October 22, 2026Link to the October 22, 2026 section
Moon CLI: a session tied to a workspaceLink to the Moon CLI: a session tied to a workspace section
On the server side, Moon AI supports the new version of Moon CLI (0.2, released October 23): a session tied to the workspace of your choice, renewed without signing in again, an authorization page that shows the device, and an email at each new sign-in (see the Moon CLI documentation).
October 14, 2026Link to the October 14, 2026 section
Link your Discord accountLink to the Link your Discord account section
On the Moon AI Discord server, type /lier: the bot sends you a link. Open it while signed in to Moon AI, check the two accounts shown, then click “Link account”. If your workspace has a paid subscription, the bot gives you the subscriber role and access to the subscribers’ channel.
In “Settings” › “Account”, the Discord section offers “Join the Discord” or, once linked, “Unlink account”.
Moon Blur: nothing is sent if the analysis failsLink to the Moon Blur: nothing is sent if the analysis fails section
If Moon Blur cannot analyze a message, nothing is sent to the model and a message tells you so; previously, the message went out without protection. The Moon Blur badge below your message also flags a partial analysis. A value that has already been replaced is no longer reread by mistake, which makes putting the real values back into answers more reliable.
The color and effect of your message bubbles can be chosen from a palette, and follow the plan of the open workspace.
See Moon Blur and Moon Blur limits.
October 8, 2026Link to the October 8, 2026 section
Created images in your documentsLink to the Created images in your documents section
The images the AI creates in a conversation are now available to the isolated computer. So you can ask for a PDF, a presentation or an illustrated document that uses these images, such as a children’s book. Photos you attached to earlier messages are not carried over: attach them to the message that asks for the document.
See Isolated computer.
October 7, 2026Link to the October 7, 2026 section
Your own data to maskLink to the Your own data to mask section
In “Settings” › “Security” › “Moon Blur”, the “Your data to mask” section lets you add your own terms: client names, project names, internal codes, addresses. Moon Blur always replaces them before sending to the model, however they are written (capitals, accents, hyphens), even in the middle of a file name, and puts them back in the answer. You can paste several lines at once; a preview tells you what will be added or rejected. The message badge shows the items “From your list”.
This list is available from the Constellation plan up (see the plans). The section also reminds you what Moon Blur does not cover: connectors, images, voice and web searches.
The AI answers questions about Moon AILink to the The AI answers questions about Moon AI section
Ask a question about Moon AI in any conversation: the AI answers with the real menu names and a link to the right page. A question about your account (your plan, what it includes, your usage, when your limit comes back) gets an answer read from your account, as a percentage.
See Support.
Archicad, Revit and IFC modelsLink to the Archicad, Revit and IFC models section
- Archicad: a guide explains how to connect a community MCP server to Archicad. The form creates the access token itself and puts it in the command to copy; any command that changes the project asks for your approval every time.
- Revit: a guide for Autodesk’s official server, read-only, from Moon CLI.
- IFC models: attach an
.ifcor.ifczipfile; the isolated computer produces a summary, quantity takeoffs, properties, an Excel or CSV export and a check.
Moon AI also refuses an Archicad or Fusion server that would answer without a token. See Design software.
A warning for very costly modelsLink to the A warning for very costly models section
In the model picker, a “Very costly” badge flags the models where a typical message uses a large share of your 5-hour limit. Hovering over it shows that estimated share.
When such a model is selected, a card above the message box says so, with a “Switch to Auto” button and a “Hide for this model” close button. It is for your information only: nothing is blocked.
See Choose a model.
October 6, 2026Link to the October 6, 2026 section
Deep researchLink to the Deep research section
Deep research is now open to everyone, depending on your plan. Choose “Deep research” in the “+” menu (or from the command palette), then ask your question.
- A card shows the expected duration, the share of your limit it will use and how many research runs you have left this month.
- The AI can ask you up to three clarifying questions and suggests a plan that you can edit before starting the research.
- The research keeps running in the background, even with the tab closed, and lets you know when it is done.
- The report cites its sources and can be downloaded as Word, PDF or Markdown.
See Deep research.
Telegram, import and notifications open to everyoneLink to the Telegram, import and notifications open to everyone section
Several features are now open to all accounts, within the limits of their plan:
- Your agents on Telegram: link your own Telegram bot to an agent and talk to it from the messaging app. The bot only listens to your account, and any action that changes something is submitted to you, with a button or a code. See Messaging apps.
- Agent triggers: an agent starts on an event from a connector, such as a new email, and reports back to you.
- Import from another AI: in “Settings” › “Data Controls”, “Import from another AI” brings in your conversations exported from ChatGPT, Claude or Gemini, with no duplicates if you run it again. You can also bring over your memory: you pick the memories to keep, one by one. See Import.
- Browser notifications: a long answer that is ready, an action to approve or a finished task lets you know, even when Moon AI is closed. They work on computers, on Android in Chrome or Firefox, and on iPhone when Moon AI is added to the Home Screen (not in the Android app). The text stays generic and never includes the content of your conversations. See Notifications.
Photos attached to your messages are now stored as files, so conversations stay light. The Library shows a “Photos of your conversations” row where you can find and delete them.
Live documentsLink to the Live documents section
“Open in editor”, in the “⋯” menu of a long answer, turns it into an editable document next to the conversation (a Markdown document created by the AI opens there too).
- Edit the text directly, in Document or Markdown mode.
- Select a passage and use “Edit with Moon”: your own instruction, “Shorter”, “Longer”, “More formal”, “Simpler” or a spelling fix. The suggestion is shown before and after, and nothing changes without your approval.
- The history keeps every version; you can restore an older one.
- Export to Markdown, Word or PDF, or copy to Google Docs.
The original answer in the conversation never changes. See Documents.
Learn with Moon: quizzes, flashcards and guided pathsLink to the Learn with Moon: quizzes, flashcards and guided paths section
Ask to review a topic or a document: Moon creates a quiz, flashcards or an interactive guided path in the panel. The “Learn” start button and the /quiz, /fiches and /parcours commands set the format.
- Quiz: answers checked after each question or at the end, hints, a score, “Retry the mistakes”.
- Flashcards: flip them, sort them into “I know it” or “Review again”, shuffle them; export them as CSV.
- Guided path: steps, each with a key point and a check question.
Your progress is saved in your browser. See Learn.
Autodesk FusionLink to the Autodesk Fusion section
A guide explains how to connect Moon AI to the MCP server built into Autodesk Fusion, on your computer, through a password-protected tunnel. The AI can then read your model, change it and run scripts in Fusion, with your approval.
See Design software.
Common actions in Slack, Jira, Linear, HubSpot…Link to the Common actions in Slack, Jira, Linear, HubSpot… section
About twenty services now give the AI the actions people ask for in a single sentence, such as sending a Slack message or creating a Jira ticket. Among them: Slack, Jira, Linear, Asana, HubSpot, Todoist, Salesforce, Stripe, Calendly, Confluence, GitLab, Intercom, Mailchimp, OneDrive, Trello, Zoom and Cal.com.
- Sensitive actions (money, mass sends, public links and sharing, permanent deletions) always ask for your approval, even with “Act without asking”; the service’s card marks them “Approval required”.
- A message going out to other people also asks for your approval when external content was read earlier in the same answer.
See Connectors.
Reaching a limit no longer leaves an empty answerLink to the Reaching a limit no longer leaves an empty answer section
When a usage limit stops a request, a notice is written in place of the answer: which limit was reached, the time it comes back, and “Continue” or “Retry” as soon as it is back. The refused question stays in the conversation.
See Usage limits.
Created videos in the LibraryLink to the Created videos in the Library section
A “Created videos” tab joins “Created images” in the Library. There you will find the videos of your active workspace, with their length and a preview, a search, and a viewer that lets you play, download, open the original conversation or delete.
Fix: each created video keeps its own file; a video can no longer replace another one created shortly before. The format chosen on the videos page is now applied.
See Library.
Images: a gallery, styles and formatsLink to the Images: a gallery, styles and formats section
- A “Created images” tab in the Library gathers the images created by the AI, grouped by date, with a search through their descriptions.
- When “Create an image” is selected, a row below the message box offers a style (realistic photo, watercolor, anime, sketch, cinematic…) and a format (square, portrait, landscape, story).
- The viewer shows the description, style and format, and lets you open the conversation, edit the image or reuse its description in a new conversation.
See Images.
Projects: conversations, sources and instructionsLink to the Projects: conversations, sources and instructions section
A project page has three tabs: Conversations, Sources and Instructions.
- Search and sort the project’s conversations, archived ones included.
- Each source shows its status (being read, unreadable, no longer available); text pasted into the project can be edited in place.
- Instructions come with examples to insert and a counter.
- A project can be renamed in place and duplicated. When you delete it, choose “Keep the conversations” or “Delete the conversations too”. A conversation leaves the project with “Remove from project”.
See Projects.
Scan a documentLink to the Scan a document section
With “Scan a document” in the “+” menu, take photos of pages: the page edges are detected and the page is flattened, then you choose a filter, the order and the pages to keep. Everything is attached to the message as a PDF, and the text is read, even on a scanned page.
See Attach files.
FixesLink to the Fixes section
- A long answer that is still being written is no longer cut off after a few minutes.
- End-of-answer notifications no longer include the thinking or tool blocks.
- A long page no longer shows two scrollbars.
October 5, 2026Link to the October 5, 2026 section
The “+” menu on phones and message box commandsLink to the The “+” menu on phones and message box commands section
- On a phone, the “+” button opens an “Add to the conversation” sheet with three large tiles (camera, photos, files), followed by the other sources and actions.
- In the message box,
/suggests actions (image, video, web search, document, skill) and your prompts;#attaches a link, a Library file, a note or a knowledge base;@picks a connector, an agent or a model. @Gmail(or another connector) has that connector used for the message. Your approval settings still apply.- CtrlShift1 opens the “+” menu.
See The message box.
Interactive answersLink to the Interactive answers section
- The AI can ask you a question with answers to click (a single choice, multiple choices or a ranking) instead of guessing.
- A small interactive visual (chart, diagram) can appear in the answer, expand to full screen, be copied as an image or be saved as an artifact.
- A web image search is shown as a gallery.
- “Verify the answer”, in the “⋯” menu, checks the claims in an answer against the web and highlights them in green or orange. The share of your limit it will use is estimated before it starts.
See Answers.
Dyslexia-friendly font and “Your month”Link to the Dyslexia-friendly font and “Your month” section
In “Settings” › “General”, turn on “Dyslexia-friendly font”: the whole interface switches to a font designed to make reading easier, with more spacing in the text you read.
A new “Your month” tab sums up the current month and the previous one in your workspace: conversations, messages sent, active days, busiest days, most used models and tools, and the share of your usage limit you used. The month’s main topics are worked out when you click, from conversation titles only.
See Settings.
Thinking level and answer style in one clickLink to the Thinking level and answer style in one click section
- In the model picker, “Thinking and style” offers a “Fast”, “Thinking” or “Deep” level for models that can reason, and an answer style (Normal, Concise, Explanatory, Formal); “Create a style” adds your own styles.
- The “⋯” menu of an answer offers “New chat from here” and “Export” (Word, PDF, Gmail draft, Google Docs).
- The regenerate menu offers “Try again with web search”.
See Answers.
Documents, skills and projects from the “+” menuLink to the Documents, skills and projects from the “+” menu section
- “Create a document” asks for a Word document, a presentation, a spreadsheet or a PDF for this message.
- “Use a skill” has a specific skill applied to your request.
- “Add to project” files the conversation in a project.
- A long pasted text becomes a card you can open, edit or remove before sending.
Images in Auto mode and meeting minutesLink to the Images in Auto mode and meeting minutes section
In Auto mode, an attached image is accepted as soon as an available model can read images: Auto picks it for you. When an audio recording is attached, a “Make meeting minutes” suggestion appears. Provider logos stay readable in dark theme.
The full voice callLink to the The full voice call section
During a voice call:
- Moon says in a few words what it is doing when a tool takes a while;
- an action to approve shows up in the call with “Deny” and “Approve”;
- you can turn on the camera or, on a computer, share your screen: an image is attached to each sentence so that Moon sees what you are showing;
- a soft sound tells you Moon is thinking (“Sound while thinking” setting);
- if the service is briefly overloaded, the question is asked again on its own.
See Voice.
Emails in Moon AI colorsLink to the Emails in Moon AI colors section
All Moon AI emails share the same layout, easy to read in dark theme and on a phone. The unsubscribe link only appears on news emails and recaps, and the unsubscribe page asks you to confirm before doing anything. The password reset link takes you straight to the code step.
File uploads: fixesLink to the File uploads: fixes section
Some files stayed in processing forever: that no longer happens. An archive lists the files it contains, a scanned PDF is read with text recognition, and a PDF with no readable text gives a clear message. A file stuck for too long is marked as failed so you can upload it again.
See Attach files.
October 4, 2026Link to the October 4, 2026 section
Agents: context, a “Pending approvals” inbox and notificationsLink to the Agents: context, a “Pending approvals” inbox and notifications section
Agents get a working context: business instructions, files and folders from the Library, chosen skills, a guided setup and a test run where actions are simulated.
- “Pending approvals”: the actions an agent wants to take while you are away wait for your decision in a dedicated inbox; you approve or deny them, then the agent picks up again. You can also set, per agent and per action, whether to always ask, always allow or forbid.
- Notifications: a bell, notifications on your phone and an email summary let you know when an action is waiting or a task is done.
- Log: each agent keeps the history of its runs, and an agent can be shared with your workspace (each member uses it with their own connectors).
- An agent’s page shows the real status of each of its tools (connected, paused, needs reconnecting), and “Connect” brings you back to the agent once the service is connected.
Routine completion notifications opened to everyone that day, with the “Notify me when it ends” checkbox. These new agents opened to some accounts first, then to everyone.
See Agents, Approve actions and Notifications.
Long conversations no longer overflowLink to the Long conversations no longer overflow section
When a conversation gets too long for the chosen model, Moon AI summarizes its beginning instead of silently cutting off the history. A line in the conversation says so, and the “Summarizing the start of the conversation…” status shows while it works. A document too long for the model is read through its relevant passages. The available room depends on each model: a model with a large memory keeps much more history.
See Conversations.
A new “+” menu and actions for a single messageLink to the A new “+” menu and actions for a single message section
The “+” menu in the message box is reorganized into groups: what you add, what you want “For this message”, and your connectors. Choosing “Create an image”, “Create a video” or “Search the web” puts a chip on the message box: the action applies to that message only. “Schedule this message…” turns it into a routine. The connector approval setting is also in this menu.
See The message box.
Limits that reset at a set timeLink to the Limits that reset at a set time section
Your usage is shown over three windows, each as a percentage: the 5-hour session, the week and the month. The 5-hour limit and the weekly limit now work in blocks: a block starts with your first request and resets fully to zero when it ends. The app shows the exact time (for example “Week · resets tomorrow at 7:56 AM” in the account menu). Before, usage freed up little by little.
See Usage limits.
Projects and creative connectorsLink to the Projects and creative connectors section
- A “Projects” section in the sidebar groups conversations with their sources and instructions. The number of projects depends on your plan.
- New connectors for creating: Figma, Pexels, official MCP servers (Unsplash, Gamma, Lucid, Whimsical, Pitch, Cloudinary, Descript, ElevenLabs), guides for Miro and Webflow, and guides for driving software on your computer (Blender, Spline, Sketch, Illustrator…).
- Fix: a created image or video no longer disappears from the conversation when the app was in the background during creation.
See Projects and Design software.
The latest models, sorted from newest to oldestLink to the The latest models, sorted from newest to oldest section
- The model picker’s home view offers Auto, Moon 6, the flagship models of OpenAI, Anthropic and Google, and a fast model.
- “All models” groups models by provider, newest first; older ones are under “Previous versions”. A “New” badge marks models less than 30 days old.
- New models arrive, including GPT-6.1 Sol, Claude Sonnet 5.5, Gemini 3.8 Flash, Qwen3.8 Max, Mistral Large 3 and Devstral 2. The Reasoning, Code and Vision filters take them into account.
See Choose a model.
Dictation, read aloud and turn-by-turn voice callsLink to the Dictation, read aloud and turn-by-turn voice calls section
Dictation and read aloud now use a new audio model. Voice calls work turn by turn: speak, pause, and Moon answers out loud as soon as its first sentence is ready. Talking over Moon pauses its voice. In “Settings” › “Audio”, adjust “End of speech” and “Interrupt Moon by speaking”.
See Voice.
Faster web searchLink to the Faster web search section
Web search answers faster. The send button comes back as soon as the answer ends, without waiting for the conversation to be saved.
See Web search.
A guided tour and “First steps”Link to the A guided tour and “First steps” section
A guided tour presents the message box, the “+” menu, Moon Blur, agents, connectors, projects and your usage, without ever blocking the screen. A “First steps” list in the account menu checks off completed tasks on its own. This tour was replaced on November 4 by the new welcome flow (see November 2026).
October 3, 2026Link to the October 3, 2026 section
Automatic memoryLink to the Automatic memory section
Moon AI can remember, from one conversation to the next, what is useful to you (your preferences, your context) and find an earlier conversation when you refer to it. Memory is off by default: you turn it on yourself after reading a notice, and you adjust it in “Settings” › “Personalization”. It is never used in a routine or a webhook, and a conversation that contains an untrusted message (for example from a community agent) is kept out of it.
See Memory.
The team spaceLink to the The team space section
In a team workspace, the “Team” page brings together what members share (conversations, notes, agents, skills, prompts, knowledge bases) and the member directory.
- A note can be shared with the workspace as read-only or editable.
- A conversation is shared with the workspace as a frozen copy that members can read; “Continue in a copy” picks it up in your own account.
- The “Share” window has two tabs: the workspace or a public link.
- A member removed from the workspace immediately loses access to what was shared.
See Team.
Follow the AI’s work step by stepLink to the Follow the AI’s work step by step section
During a task, the AI says in one short sentence what it is about to do, in the language of your message. Its steps are grouped on a gray line you can expand (for example “Searched the web, created 2 files”), with a title for each step and failures flagged separately. Sources appear as chips with the site’s icon and name.
See Answers.
Images and videos, full sizeLink to the Images and videos, full size section
The tools wrench leaves the message box: the AI decides on its own when to search the web or create an image. An image being created shows up in a frame with the right aspect ratio, then appears full size with “Download”, “Copy” and “Edit”; a video arrives with its player. A refusal from the model or a limit being reached is explained in a card.
Agathe also handles your Outlook appointmentsLink to the Agathe also handles your Outlook appointments section
Agathe, the appointment scheduling agent, can now look up, change and cancel an event in an Outlook calendar, always after your approval. When a task or a routine stops because of the plan, a “See plans” button appears for the workspace owner.
Edit an agent’s toolsLink to the Edit an agent’s tools section
The “Edit tools” button adds or removes connectors on an agent you have already created.
Fixes: resuming after an approval works with recent Claude models; an answer interrupted after using tools offers “Resume” instead of starting over; the skills and memory toggles keep their position; dropdown lists now show only one arrow.
The team directoryLink to the The team directory section
In a team workspace, members’ email addresses are only visible to the owner and administrators; other members see names. Account deletion now makes clear that copies made by your colleagues stay with them.
See Members.
FixesLink to the Fixes section
- Conversation titles and tags are written in the language of the conversation.
- Text that comes from a link (for example “Find a slot”) is no longer kept as a draft if you didn’t change it.
- Moon’s crescent now only appears on Moon AI models.
- Several texts that were cut off on phones are fixed.
October 2, 2026Link to the October 2, 2026 section
The routines calendar and the scheduling agentLink to the The routines calendar and the scheduling agent section
The routines page switches between “List” and “Calendar”: month, week, day or agenda views (agenda by default on a phone). On a computer, you move a one-time routine by dragging it, and an empty slot lets you create one. If you have connected Google Calendar or Outlook, your events show up next to your routines.
Agathe, the appointment scheduling agent, looks for a free slot in your calendars and prepares the invitation; she always asks for your approval before writing to a calendar. The calendar’s “Find a slot” button opens her directly.
See Routines.
A redesigned developer API pageLink to the A redesigned developer API page section
The developer API page is organized in tabs: Overview, Usage, API keys, Billing, Models and Limits. The overview shows your spending for the month and the day and the status of your caps; the wallet history and the API address are displayed.
See Developer API.
Files below the answer, and always a written answerLink to the Files below the answer, and always a written answer section
- The AI always ends with a written answer, even after using tools.
- The files it creates appear below the text, as cards with a preview or a download; images are laid out in a grid.
- In Agent mode, a card shows the current step, the progress and the time spent; a “Continue” button picks the work back up when the turn limit is reached.
A floating message boxLink to the A floating message box section
The message box floats above the conversation, which scrolls underneath. A request to approve an action stays visible with its “Deny” and “Approve” buttons; “Don’t ask again for Gmail” (or another service) asks for confirmation and can be undone. Below the message box, the notice “Moon AI is an AI and can make mistakes.” is followed by a “Moon Blur” button that explains the protection.
See Approve actions.
Settings: security, data and workspaceLink to the Settings: security, data and workspace section
- Security: signing out now only closes the current device; each signed-in device can be signed out separately, and “Sign out other devices” closes all the others. Backup codes work for signing in. An account created with Google or Apple can “Set a password”. “Recent activity” lists the account’s security events.
- Data Controls: “Shared links” are listed and can be removed; archiving, deleting and exporting apply to the open workspace.
- Workspace: the page is organized in tabs, and all members see the workspace’s plan.
Clean copy and paste, and Word documentsLink to the Clean copy and paste, and Word documents section
An answer copied into Word, Google Docs or Outlook keeps its formatting, but no longer the theme’s colors and backgrounds (pasted text is no longer light gray). Clicking a Word or Excel document created by the AI downloads it directly. The “Download” buttons in preview windows work again.
Open the sidebar with a swipe on your phoneLink to the Open the sidebar with a swipe on your phone section
On a phone, the sidebar opens over the conversation and follows your finger: swipe from the conversation to open it, and the other way to close it.
New pages flaggedLink to the New pages flagged section
A “New” badge flags the recent pages in the sidebar (Agents, Pending approvals, Routines, Connectors, Skills) until your first visit. The old action that summarized the conversation is removed, and the greeting on the home screen fits on one line.
Sign-in: no more codes asked for no reasonLink to the Sign-in: no more codes asked for no reason section
Once your email address is verified, signing in no longer asks you for a verification code again.
October 1, 2026Link to the October 1, 2026 section
About forty new connectorsLink to the About forty new connectors section
The connector catalog grows: CRM and prospecting, customer support, e-commerce, marketing and email campaigns, human resources, analytics and SEO, developer tools. Five categories join the “Connectors” page, and each service’s page adds a “Before you connect” section that tells you what you will be asked for. Some sensitive actions always ask for your approval, even in “Act without asking”.
See Connectors.
The plan that unlocks an optionLink to the The plan that unlocks an option section
Wherever an option is locked, a badge names the cheapest plan that unlocks it, and the plans window only offers the plans that do. The same evening, what each plan includes was revised: Lune includes connectors and routines, and some plans come with a monthly amount to spend on the developer API. The plan cards describe what is included. The Free plan for accounts created from that day on gives access to the economy and standard models. The details are on the pricing page.
Developer API: OpenAI-format errors and Moon BlurLink to the Developer API: OpenAI-format errors and Moon Blur section
Developer API errors now follow the OpenAI format, with a clear code (invalid key, insufficient balance, unknown model, rate limit). Responses no longer carry the provider’s internal fields. Moon Blur also applies to API requests, with a setting per key, and a response header tells you whether it was applied; files, images and audio you send are not masked.
The command paletteLink to the The command palette section
CtrlK (⌘K on Mac) opens the command palette from anywhere in the app. It finds an action (new chat, note, agent, theme…), a page, a conversation, an agent or a file from the Library, and ignores accents.
See The interface.
Your email inbox through IMAP and SMTPLink to the Your email inbox through IMAP and SMTP section
Connect an inbox from most providers (OVH, Infomaniak, Gandi, iCloud, Gmail with an app password…) through its IMAP and SMTP servers. The AI can search and read your emails, save an attachment to the Library, mark as read, move, reply and send.
- Every action on the inbox asks for your approval by default; sending from a routine always goes through “Pending approvals”.
- An email the AI reads is treated as content written by someone else: for the rest of the conversation, every action asks for your approval.
- For Outlook and Microsoft 365, use the Outlook connector.
See Email inbox.
Redesigned sidebar and menusLink to the Redesigned sidebar and menus section
The sidebar keeps the essentials (new chat, search, Agents, Library, Notes) and moves the rest under “More” (Pending approvals, Routines, Connectors, Skills, Studio). “Help and support” moves to the account menu, which also shows a “Usage” line. Fix: a window opened a second time responds to clicks again.