YouTube summaries: Drop a YouTube link and your agent watches it with Gemini, then summarizes it or answers your question about it.

Model fallback: Your automations and chats no longer stop when a model is unavailable. Atmita now automatically falls back to another model, so work keeps running. You can turn this off in Settings.

We renamed the Wiki app to Library. We also merged the Skills app into it, so it's no longer a separate app. The Library now has two tabs: Wiki and Skills.

Claude Fable 5 is now a model option. Anthropic's most capable model yet. Point it at your hardest automations and missions.

When an automation or mission produces something worth keeping, it now adds it to your wiki on its own. You can turn this on or off for each one.

Agents can verify that an email address is real and deliverable before sending to it.

Wiki pages can now be shared with other people, either read-only or with edit access. Each page shows who it is shared with, and collaborators can edit it in place.

You can search across your past chats. The agent can look back through every earlier conversation by keyword, narrow by date, and bring an old chat back into view.

Change previews now show old and new versions side by side, with the exact words that changed highlighted.

The Wiki became a full surface of its own, with its own layout and a suggest button.

Missions now show in your calendar, and you can wake one immediately with a note to steer its next round.

New side-by-side pages show how Atmita compares against fifteen other tools, including when each one is the better fit.

A refreshed look across the site, with a new logo, cleaner type, and a darker background.

Developer stats and extra apps no longer clutter the dashboard by default; long-press to reveal them.

Missions arrived: a goal that runs over many rounds, adapts as things change, and can pause to ask you a question.

Agents can securely look up and use logins you've saved when a task needs them.

Plan Mode lets an agent think through and lay out its plan before taking any action.

You can see and edit an automation's memory right from its card, so you can adjust what it remembers.

The Notes app was retired from the main lineup. It still exists, turned off by default, and you can bring it back from the tools section in Settings.

You can connect Twitter and have your agents post and work with it.

The connected-apps section in Settings was removed; connections are now managed right in chat.

Agents can read full PDFs directly from a link.

Agents can open and use a real web browser when a task can't be done any other way.

Math now renders cleanly in your messages, with centered, readable equations.

You can connect your own apps and tools to Atmita and have it use them on your behalf, with a search to find what you need.

You can pin any chat to the top of your history.

Scheduled automations can now skip a run entirely when there's nothing for them to do.

You can browse your memory history, edit any entry, and delete the ones you don't want.

You can create images inside chat and attach them to your messages.

You can pick which agent you're talking to from the chat box and switch agents mid-conversation.

Arabic is now fully supported, with right-to-left layout and formatting across the app.

A "What's New" page now lives in the app so you can follow the latest updates.

You can pick which AI model powers your chat directly from the message box.

You can add images to chat by pasting or dragging them in, and open any of them at full size.

A confirmation step now appears before anything is deleted or undone.

Automations can now react to outside events and run only when the event matches what you set.

The Mail app was removed as a standalone app. Email later returned as a built-in capability agents use to read, write, and send, without a dedicated mail screen.

The full-screen chat was removed from inside apps; chat now opens in a split pane beside what you're doing.

The Memos app was retired, replaced by the new Notes app.

A new Notes app organizes everything as a tree, where any note can hold tasks, projects, and tags beneath it.

A preview card shows the assistant's latest reply over the app you're in, so you don't have to switch to the full chat.

Apps generated on the fly by the AI were retired. Atmita moved to a set of solid built-in surfaces instead of generating a new app for each request — more reliable, at the cost of free-form app building.

Agents can search the web live to look things up while they work.

Setting up an automation is simpler, with a separate "when this happens" trigger and "do this" action, and scheduling that understands plain-language weekdays and your local timezone.

Sensitive actions now pause and wait for your approval before they run.

Four agent modes let you set how an agent acts: Safe asks before every action, Auto runs on its own, Bypass skips approvals, and Plan only drafts a plan without doing anything.

Automations can now be triggered by incoming mail, so a new message can start an action on its own.

You can ask the AI to draft an email, then review it and send it in two steps.

A built-in Mail app arrived, with your inbox, sent, drafts, and trash in one place, and your Gmail and Outlook syncing both ways.

The manual "create app" flow was retired; the AI handles app creation end to end.

Any change the AI makes can be kept or undone, one at a time or all at once, so nothing is permanent until you say so.

The AI plans its work before building, edits only the part you ask about, and fixes its own errors as it goes.

Atmita launched. You could describe an app in plain words and the AI would build it for you, running either in the cloud or directly on your device.