Skip to content
All changes

Changelog

26 October – 1 November 2026

11 new · 8 improved · 15 fixed

New

Ask the agent about your plan, credits and sites

The agent can now read your own account. Ask "How many credits do I have left?" or "What plan am I on?" and it answers with your balance, your plan, your published sites and which top-up packs you can buy.

It also knows Automatio itself much better: plans, prices, top-up packs, the own-key plan, voice calls, Telegram and where things are on screen. Quick questions like these are answered right away even while a longer task keeps running.

Learn more
New

The agent can point at the right button on your screen

Ask "how do I publish this?" or "where are my account settings?" and the agent can put a ring and a short note on the real button, instead of only describing where it is.

For the account menu, the agent panel, the model picker, the project panel and the Publish dialog, it can open them for you. It can also point at the voice call button.

New

Public docs, now in five languages

Automatio now has public docs at automatio.ai/docs: 22 pages on how the product works, from chats, models and memory to Canvas, automations, the browser, publishing, plans and privacy.

The docs are also available in Russian, Chinese, Korean and Indonesian, with a language menu in the sidebar, and each page has a card that takes you to our Discord.

Learn more
New

Canvas workers wait for each other and hand over their full work

On a Canvas, a worker told to wait for others now starts only when all of them have finished, instead of starting on half-done work.

The next worker now gets what the previous one actually produced, including the documents it made, and the master chat reads what each worker handed in, so it can tell you when a build does not match its design. The master can now also draw charts, tables and other interfaces in its answers.

Autonomy and budget now sit together in one settings menu on the board and on your phone.

Learn more
New

Notifications: a bell, a log and push to your phone or computer

A new bell in the sidebar and a notifications page tell you what happened while you were away: a task finished or failed, a task stopped to ask you something, an automation's result, or a reply to your bug report.

  • Push notifications work with the app closed, in Chrome, Edge, Firefox and Safari (on iPhone, add the app to your Home Screen first).
  • On Windows and Android, a plan's notification has an Approve button.
  • Bug reports are now a conversation: you and our team can write to each other on the report.
  • On Android, tapping a notification opens the chat, and the status bar icon shows a clean "A".
  • "Send a test" now shows on the device you pressed it on and tells you, by name, which devices got it.
Learn more
New

A second request sent mid-task gets its own answer

If you send a message about something different while the agent is still working, it now waits its turn. The agent finishes what it was doing, answers that, then picks up your new message and gives it a reply of its own, so neither request gets lost.

A waiting message shows a "Queued, next after this task" pill, and the pill stays there even if you reload the page, so you can see it was heard and do not need to send it again.

Learn more
New

Your Discord helper answers from a real knowledge base

Ask your own agent something like "make a knowledge base for my server from our website, our docs and this PDF", and the Discord helper will answer members' questions from it. Changes to the knowledge base wait for your Approve, like every other change to what the helper tells members.

The helper is also better company:

  • It reads the room and stays quiet when people are talking to each other.
  • It knows your server's role rules from its live settings and can tell members how to get each role.
  • It knows which channels it listens in.
  • Members can use /privacy to stop the bot reading their messages, turn it back on, or delete what it kept.
  • Invite links, the home page and pricing show as Discord cards with buttons.
Learn more
New

Apps you build can read video and PDFs, use tools and edit images

Apps built on Automatio can do more through the platform API:

  • Send a video link or a PDF to the chat and have it read.
  • Use tools, JSON mode, stop words, seed and the penalty settings that apps written for the OpenAI SDK expect. Anything not supported is refused by name instead of silently ignored.
  • Generate images from up to 4 starting pictures and ask for 1 to 4 images at once, so an app can edit a picture or keep the same character across frames.
Learn more
New

See what your work cost in dollars, at your own prices

Reply and chat cost popovers now show an estimate in dollars first, worked out at your own price per credit, with the credits underneath. Where several models were used, you see a line for each, plus tools.

The browser batch progress card shows the same: its own credits with the dollar estimate, the rows done and the real run time, including after a batch has finished.

Learn more
New

New models: GPT-6.1 Terra, Particle-1 and four more

Six new models are available in the model picker, including GPT-6.1 Terra and Particle-1, plus four other models released in October.

Claude Sonnet 4 is no longer offered. Chats and saved preferences that used it now move to Claude Sonnet 4.5.

Learn more
New

Discover ready-made app connections in the MCP tab

The MCP tab now opens on "Discover servers": 29 servers run by the companies that make them, with logos, categories and a search box, so you can find a connection without hunting for an address.

When you add a server, the dialog asks it how it signs in and picks the right method for you. Deleting a server now asks first. When you ask the agent to find a server in chat, it searches this directory first and tells you plainly if the public list did not answer.

Learn more

Improvements

  • ImprovedYour phone screen stays on while you dictate, so a long recording is not cut off when the screen would lock.
  • ImprovedThe skill viewer has View and History tabs and works properly on a phone screen.
  • ImprovedThe Publish dialog explains why production was built fresh instead of reusing the staging build.
  • ImprovedSearching your past chats finds a chat by the words inside it, not just its title.
  • ImprovedThe use-my-Chrome card offers to install the extension when this browser cannot take commands yet.
  • ImprovedBatch panels show how many jobs completed while one of their steps failed.
  • ImprovedThe agent sees every column of tables from Google Analytics, Search Console, Sheets and other data tools.
  • ImprovedCreating a Discord channel inside a category now works in one step.

Fixes

  • FixedLong Claude runs finish instead of stopping with an error partway through.
  • FixedOn your own Anthropic key, a run keeps going when the provider declines one step.
  • FixedOn your own key, a refused request says your connection works and that your next message continues.
  • FixedForms with fields drawn by the agent show up again.
  • FixedIn the installed phone app, swiping from the left edge opens the sidebar instead of going back.
  • FixedRestoring an earlier version of a project puts its matching packages back, so the preview runs again.
  • FixedPublishing goes through even when the project's server is busy.
  • FixedThe Discord helper no longer treats an old screenshot as part of a new question.
  • FixedWhen you decline the Discord helper's own plan, the decline stays on your card and is not posted for members.
  • FixedA chat too big for its model shows a short notice in seconds instead of thinking for about half an hour.
  • FixedChats with large phone photos keep working on later turns.
  • FixedWhen a run fails, the reply you get knows what the run had already done.
  • FixedThe agent tells you the right number of apps you have, even when you have many.
  • FixedThe coding agent's code map tools work on big repositories.
  • FixedDaily Discord role checks for pictures and X profiles run again.