Twin 1.3.2
- The agents window opens joined to the notch, as Twin's own window does, with no gap between them, whether you open it from Expand, the notch, the menu bar or an agent's face.
Changelog
Every Twin release
Every release, newest first, with the same notes the update shows you. Mac and Windows ship together.
- Latest
- 1.3.2
- Released
- 7 October 2026
- Releases
- 6
- 1.3.2
build 7
- 1.3.1
build 6
Twin 1.3.1
- Agents that specialise
- Every agent now knows Twin's skills for files, websites and its own computer, research, your apps and meeting notes, and uses the ones its job needs.
- An everyday job gets an agent like before; a specialised one gets a specialist named for it, like "Trip Twin" or "Invoice Twin".
- Agents wear six new faces, everywhere they show: the dock, the agents window and the notch.
- Quicker, surer decisions
- Twin decides what you're asking for in about half a second, before its answer is even written: whether to tell you, show you how, do it for you, search the web or hand it to an agent. It now sees what's on your screen and remembers what you just asked when it decides, so "do it" and "send it to him" go where you mean.
- Anything that changes something on your computer happens only when Twin is sure that's what you asked for; otherwise it offers it first, with Do it and No.
- Questions about the news or anything current search straight away, so their answers come sooner.
- "Say that again" repeats the last answer at once.
- On Windows
- Tasks Twin does on your screen now run step by step instead of stopping at the first step, and an action you say yes to carries on where it left off.
- Agents' computers, sturdier
- An agent's computer now starts reliably, even when its first start takes a minute or two, and agents can take on long jobs: many steps on a site, downloads, and commands that run in the background while they check in.
- New agents can use their computer's files and terminal straight away. Switch any of it off in the agent's Computer panel.
- 1.3
build 5
Twin 1.3
- A new look
- Twin has a new icon: a little platinum screen with the sky, the hills and Twin's sparkle, and a ^ on its chin.
- Your apps, properly
- Ask about your email, calendar or other connected apps and Twin works through them step by step, reading first and then acting on what it found ("find a free half hour tomorrow and book it with Ravi").
- You can watch each step in the notch (on Windows, in the pill) as it happens.
- Anything that sends, books, creates or deletes waits for your yes. Twin asks in its own words, and a yes runs it exactly once.
- Hears any language, answers seeing your screen
- Twin now hears you in any language, even when you switch mid-sentence, and answers by looking at your screen itself: it points at icons as well as words, and starts talking sooner.
- Twin answers in your language too: Settings › General › Language now has Spanish, French, German, Italian, Portuguese, Dutch, Japanese and Russian, and Match how I speak answers in whichever of them you spoke, out loud in that language.
- Agents, notes, tasks and dictation write with the same model as Twin's answers, worldwide, and Sarvam for Indian languages.
- New voices: ten natural voices (American, British, Australian) speak English and every other language, and the Indian voices speak Indian languages, which they say best. Settings › Voice has a picker for each.
- Settings › Microphone › Language you speak: leave it on Any language, or pick Indian languages (or one of them) to have Sarvam listen and answer.
- Twin gets to know you
- The first run now asks you a few quick questions: what you do, what you're working on, the apps you live in, and what you'd love off your plate. Type or hold ⌃⌥ and say it, or skip any of them.
- With memory on, Twin remembers what matters from your answers, then offers a few helpers made for you. Keep the ones you like, and start one right away.
- "What's good, Twin?"
- Settings › Keys › Wake Twin by voice: say "What's good, Twin" (or "Hey Twin") and ask, no keys needed. It's heard on your computer only. Say just the phrase and Twin answers, then listens for what you want.
- Two Gmails? Both work
- Connect more than one account of an app (work and personal), name them, and tell Twin which to use: "email Ravi from my work Gmail". Settings › Connections.
- Pictures
- Ask Twin to draw something, or to explain something with a picture, and it makes the picture and shows it in the notch, saved in Pictures › Twin. Circle a picture and say "make this a cartoon" to remake it.
- You decide how much Twin does
- Settings › Privacy › Let Twin use your computer: turn it off and Twin only points and tells you what to click or type; it never clicks, types or presses keys for you.
- How hard Twin thinks
- Settings › General › Thinking: High, Very High or Max. Higher levels think longer for work with several steps.
- Agents
- Every agent in the cloud now has a computer of its own on Twin's server: a browser that remembers where you signed in, a folder for its files and a terminal. Agents look things up with a normal web search and open their browser only for hands-on work, like clicking through a site or checking a live price.
- Watch it live: the agent's chat shows what its browser shows, as it works. Open its Computer panel to start or stop it, choose what it may use (Browser, Files, Terminal), and take over to sign in or click something yourself, then hand it back.
- Agents never buy, send, submit or type a password on their own: they leave those to you, and tell you where.
- Agents in the cloud use your apps the same way.
- Stop now really stops an agent's work, and the agent stays, so you can carry on with it.
- A follow-up on a file an agent made, like converting it again, goes to that same agent.
- Each agent keeps what it makes in its own folder in Documents › Twin, named after it. Open it from the agent's ⋯ menu.
- When an agent wants to act in one of your apps, answer No, Yes, or Always for this agent: then it acts in that app without asking again (never for payments, or its own browser). Stop allowing it from the ⋯ menu.
- Edit what agents write
- A document an agent makes opens in an editor beside its chat. Change anything, and it saves as you type; ask the agent for changes and it works from your edited version. Export a PDF whenever you like.
- Coding agents in the notch (Mac)
- Using Claude Code or Codex? Turn on Settings › Tools › Coding Agents, and when one asks for permission or asks you a question, answer it from the notch: Allow, Deny or Always, or hold ⌃⌥ and say yes or no. Twin shows you exactly what will run and changes your settings files only after showing you the change.
- Tell us what you think
- A new Feedback page in settings: send feedback, ask for a feature or report a bug, and the team gets it straight away.
- Fixes
- Signing in can't get stuck any more: Cancel always works, Try Again appears if the page hangs, and a sign-in that never answers gives up on its own.
- A follow-up to an agent you stopped works again.Windows
- 1.2.1
build 4
What's new in Twin 1.2
- Resize Twin's notch window by its edges; pop Home out into its own window
- Quick peek on the notch, unread badge, pin, archive, drafts
- Routines, research agents with sources, ideas from what Twin remembers
- Dictation: hands-free double tap, personal dictionary, safe in terminals
- Sounds, agents dock, the same tour as the Mac, no more flickerWindows
- Sign in with Google or Facebook
- Faster answers
- 1.2
build 3
What's new in Twin 1.2
- Pop Home out into its own window, and resize the notch window by its edges
- Routines, research agents, and ideas from what Twin remembers
- Hands-free dictation and a personal dictionary
- Little 2000s sounds, and faster answers
- 1.1
build 2
Twin 1.1
No notes were published with this release.
Every file is also on GitHub Releases.