September 30, 2026Link to the September 30, 2026 section
The Routines page and webhooksLink to the The Routines page and webhooks section
A “Routines” page brings together your workspace’s scheduled tasks. Filters separate active routines, paused ones, those that need attention and those created by the AI.
- A routine’s page shows its run history, its schedule with the next computed dates, and the share of your limit each run used.
- “Run now” starts a routine right away.
- A routine can be triggered by a webhook: an address and a secret key, shown only once, that your tools call. The content received is treated as external data, and a run started by a webhook never acts on your connectors.
- A routine created by the AI cannot write to your connectors until you have allowed it, and a routine that reaches your limit several times pauses itself.
The number of routines and webhooks depends on your plan. See Routines and Webhooks.
Skills and the isolated computer for everyoneLink to the Skills and the isolated computer for everyone section
The isolated computer and skills are now open to everyone, depending on your plan. A skill is know-how that the AI loads when a task calls for it. Twenty Moon skills are available, including: Word, PDF, PowerPoint and Excel documents, charts, data analysis, invoices, diagrams, meeting minutes, business plans, translation, long-form writing, résumés, web pages, sales preparation, and a skill that helps you create your own. The AI can also offer to save a method it has just applied as a personal skill; nothing is saved without your consent.
Fix: your data export once again contains all of its sections, including your personal skills.
See Skills and Isolated computer.
ArtifactsLink to the Artifacts section
The AI can create a page, a small app or a document that opens in a panel next to the conversation. Each change creates a new version, which you can find in the panel. Switch between the preview and the code, full screen if needed, and “Fix” sends an error back to the AI so it can repair it. An artifact stays visible in a shared or copied conversation.
Artifacts run in an isolated frame, with no network access. When the conversation contains content from outside (a web page, a file, a tool result), you are asked to click before the artifact runs.
See Create.
A page for your connectorsLink to the A page for your connectors section
A “Connectors” page joins the sidebar, right after “Agents”. It brings together the services you have connected, the full catalog and your MCP servers.
- Search ignores accents, and filters sort the catalog by category.
- A service’s card separates what the AI can read from what it can change, and reminds you that Moon Blur does not apply to third-party services.
- Each connector can have its own approval setting before it acts.
See Connectors.
Short animationsLink to the Short animations section
Dialogs, menus, tabs and the sidebar open and close with short animations. On phones, a sheet closes by following your finger. Loading placeholders no longer flash on screen for a split second. If your device asks for reduced motion, nothing moves.
Workflows make way for agents and routinesLink to the Workflows make way for agents and routines section
The old workflow builder is no longer available. To automate a task, use routines and agents.
September 29, 2026Link to the September 29, 2026 section
AgentsLink to the Agents section
The “Automation” entry becomes “Agents”. An agent is a specialized assistant, with its own instructions, connectors and starter suggestions.
- Install a ready-to-use Moon agent, create your own, or copy an agent published by another member of the community.
- A new conversation with an autonomous agent starts directly in Agent mode.
- A community agent follows stricter rules: its actions on your connectors always ask for your approval, and nothing remote is loaded in its answers until you click.
- If an agent’s model is no longer available, the “Repair the agent” button gets it working again.
See Agents.
A consistent interfaceLink to the A consistent interface section
Every screen uses the same buttons, fields, menus and dialogs. On phones, dialogs open as a sheet from the bottom of the screen.
- The workspace and account menus show your avatar, your plan and a ring that shows your usage for the month.
- After a refusal caused by a limit, your question goes back into the message box.
- On phones, the “+” menu gives access to connectors.
September 28, 2026Link to the September 28, 2026 section
Connector permissions and routine daysLink to the Connector permissions and routine days section
Below the message box, the “Connector permissions” menu offers two choices: “Approve each action” or “Act without asking”.
Fix: a routine scheduled from Monday to Friday was running from Tuesday to Saturday. Days of the week are now respected. A routine whose model no longer exists uses the default model instead of failing.
September 27, 2026Link to the September 27, 2026 section
Clearer billing emailsLink to the Clearer billing emails section
The workspace owner and its billing contact now receive:
- a confirmation for every cancellation request and when a subscription ends;
- a reminder as soon as the first payment is declined, with a link to the invoice;
- a reminder three days before a trial ends, with the actual amount;
- for an annual subscription, a notice before each renewal.
Each seat added to a workspace now gets a discount, as announced on the pricing page.
September 26, 2026Link to the September 26, 2026 section
Change plans and see the amount firstLink to the Change plans and see the amount first section
On the “Billing” page, a plan change first shows what will be charged right away and the amount of the next renewal, then asks for your confirmation. A workspace can no longer have two subscriptions at once: double-clicking the payment button no longer creates a second subscription. If your bank asks for verification, the invoice opens so that you can confirm it.
September 25, 2026Link to the September 25, 2026 section
MCP servers with sign-inLink to the MCP servers with sign-in section
MCP servers that ask you to sign in to your account (Notion, Linear, Atlassian…) are added with the same form: choose “Sign in (OAuth)”, sign in with the service, then come back to Moon AI. If the sign-in expires, the server shows that you need to reconnect.
See MCP servers.
Link previews below answersLink to the Link previews below answers section
When an answer cites articles, products or videos, cards appear below it with the title, the image and the site. In a conversation that contains private data (files, connectors, memory), no page is opened to build these previews.
The time your limit comes backLink to the The time your limit comes back section
When a usage limit is reached, the message tells you the time you can continue (for example “at 2:32 PM” or “tomorrow at …”). The plans window now only offers the plans above yours. The sidebar remembers whether it was collapsed from one visit to the next and shows your usage percentage.
See Usage limits.
September 24, 2026Link to the September 24, 2026 section
Choosing the model in the message boxLink to the Choosing the model in the message box section
The model picker is now in the message box. The home screen greets you based on the time of day and offers starter buttons (write, learn, code, brainstorm ideas, search the web, create an image).
Below each answer, the common actions stay visible (copy, like or dislike, regenerate) and the others move to the “⋯” menu. The regenerate menu offers “Simpler”, “More professional” and “Try another model”.
See Choose a model.
Your own MCP serversLink to the Your own MCP servers section
With “Add an MCP server”, connect a remote MCP server with its address and, if needed, an access token. The test button lists its tools before you save. The server’s tools are then offered to the model in your conversations; an action that changes something asks for your approval, unless you trust that server. Guides explain how to connect FreeCAD and CadQuery.
See MCP servers.
Approve the AI’s actions on your servicesLink to the Approve the AI’s actions on your services section
Before sending an email, creating a ticket or changing a connected service, the AI shows you a request above the message box: the service, the action and its main parameters. You accept or decline it, then the answer continues with the actual result.
See Approve actions.
September 23, 2026Link to the September 23, 2026 section
Normal, Auto and Agent modesLink to the Normal, Auto and Agent modes section
- Auto picks the right model for each request. A badge next to the model name shows which one answered and why. If that model fails before answering, another one takes over.
- Agent starts with a plan of a few steps, checks them off as it works and makes sure it has not missed any before stopping.
- The mode you choose is saved with the conversation.
Below your message, a Moon Blur badge shows what was masked before sending to the model, with two views: “Your message” and “Sent to the model”. The AI’s work steps are grouped in a collapsible block, and sources open in a panel on the right.