noHuman Team 4.5: browsers that say what happened and ask you for help

noHuman Team 4.5 changes how your noHumans use their browsers. A noHuman keeps track of its tabs, knows whether a failed click went through, and asks you with a card when a site needs a person. noHumans that build apps can read a page's console and network, and any noHuman can record its browser as a video.

Status: new in noHuman Team 4.5.0, released on 26 October 2026. The Mac installers come first, and the Windows 10/11 installer follows.

Install 4.5.0, read what's new in 4.5.0 or see the release notes on GitHub.

What changed in noHuman Team 4.5?

• Tabs: a noHuman works in its own tab, follows a link that opens a new one, and moves to the right tab when one closes or crashes.
• Failed steps: a noHuman learns whether a failed click or typed entry did nothing, went through, or can't be told apart, and reads typed text back.
• Browser help cards: when a site needs you, you get a card on the computer and the phone that closes itself once the step is done.
• Tools for builders: console messages, network requests with secrets removed, blocked or test responses, resent requests and a timeline of each action.
• Video, tables and full pages: an MP4 recording of the browser, web tables as rows and columns, and full-page screenshots.
• Page reading, charts and a fix: better reading of dialogs and embedded frames, chart and diagram tools, and two teams that no longer get mixed up.

The browser tools belong to the app, so they work the same whichever model a noHuman runs on: Claude, ChatGPT or another model through OpenCode.

How does a noHuman keep track of its browser tabs?

Each noHuman knows which tab is its own and works there. When it clicks a link that opens a new tab, it follows that tab. When its tab closes or crashes, it moves to the tab that opened it, or the one it used last, and is told it moved, so it reads the page again before clicking anything.

A noHuman can now list its tabs, switch between them, open a new one and close one. It never replaces a page it didn't pick. If a tab became current on its own, say after the noHuman's tab closed, the noHuman opens a new tab before going somewhere else.

What happens when a click or a typed entry fails?

The noHuman learns which of three things happened: nothing was done, the step went through, or the app can't tell. In the last case it looks at the page before it acts again, so a post, a message or a payment isn't sent twice because the first try only looked like it failed. Each failure also says what to do next.

Typed text is read back from the field. If the field cut the text or refused it, the noHuman hears so. If the field only reformatted it, such as a phone number with spaces added, the noHuman is told that this is fine and it shouldn't type again.

How does a noHuman ask me for help in its browser?

With a Browser help card. Some steps need a person: a sign-in, a captcha, a 2-step code, a payment or consent screen. The noHuman says what to do, and the card reaches you on the computer and in the mobile companion, like any question a noHuman asks you.

You open the noHuman's Browser panel and do the step there. The noHuman tells the app what the page looks like once you are done, such as an address or a button that appears, and the card closes by itself when the page shows it. You can also press Done, or I can't do it now, and the noHuman carries on without that step. A card waits 15 minutes by default and at most an hour.

While a help card waits, or you have control of the noHuman's browser, its network log keeps no contents of what the page sends. Sign-in pages never have their contents kept.

Take control is now the only thing that pauses a noHuman in its browser. In 4.3 and 4.4, Show window also paused it until you pressed Done signing in, even when you only wanted the picture of a minimised window back. To sign in in the real window with a password manager or a passkey, take control first, and the noHuman waits while you type.

What can noHumans that build apps see?

The things a developer opens the browser's developer tools for. A noHuman can read the page's console messages and errors, and its network requests with their status and timing. Passwords, cookies, tokens and similar values are removed before the noHuman sees them. It can ask which requests were slow or sent more than once.

It can also change traffic to test a fix: block a request, answer it with test data, or change it before it goes out. These rules pause during uploads and downloads and end with each task. It can send a request again from the page, to the same site only. A timeline shows, for each of its recent actions, the requests and console errors that followed.

If you have blocked clicking, typing or uploading in a browser under Safety → Forbidden, the network rules and resent requests are blocked too. Reading the console and the network stays allowed.

Can a noHuman record a video of its browser?

Yes. A noHuman can record its browser as an MP4 video without sound, 2 minutes by default and up to 10 minutes. The file goes into the recordings folder of its workspace, next to its downloads. Recording needs ffmpeg on the computer; on a Mac with Homebrew, brew install ffmpeg adds it. Without ffmpeg, the noHuman is told that nothing was recorded.

Two smaller capture tools came with it. A noHuman can read a table on a page as rows and columns with their headers, including merged cells, and it says when it saw only part of a long table. A full-page screenshot now loads the whole page first and shows a sticky header once, instead of repeating it down the image.

What else changed in 4.5?

• Better page reading. An open dialog hides the page behind it, a banner covering a button is named, clickable boxes are found, and sign-in and payment frames embedded from other sites are read. Before each click, the noHuman checks that the item is still the one it saw. Password and card number fields show as dots in what it reads.
• Charts when you ask. noHumans now have a chart tool and a diagram tool, so a chart you ask for is drawn right in the chat, as 4.4 chat blocks are, instead of arriving as an image file.
• Two teams at once. With two teams running, a noHuman could end up working inside a noHuman of the other team and send messages under its name, because both had been given the same port for their AI software. Each noHuman now checks that the AI software it talks to is its own, and the app hands out ports for all teams together.
• Shorter browser instructions. The browser guidance each noHuman reads every turn is a third shorter, and it looks up the details when it needs them.

How do I update to 4.5, and do I keep my team?

Choose Check for Updates in the app's menu, or install 4.5 from the install page. Your teams, conversations, Memory and workspace stay outside the app bundle, so replacing the app keeps them. The app stops a running team before an in-app update installs, so finish or pause work first.

Quick answers

Does a noHuman see my password when I sign in for it?

It isn't meant to. Password and card number fields show as dots in what it reads, and its network log keeps no contents while you have control or a help card waits. These limits cover how the noHuman reads pages. They are not a sandbox, so give a noHuman's browser only the accounts you want it to use.

Do the new browser tools work with ChatGPT?

Yes. They are the app's own tools and work with any model a noHuman runs on.

Is noHuman Team 4.5 out for Windows?

4.5 ships for Mac first, and the Windows installer follows. Until then, Windows users can run 4.4.0.

Can I see a help card on my phone?

Yes. The mobile companion shows it with your other questions. Its iOS and Android builds are prepared for store release, with no public download yet.

Install noHuman Team to give your noHumans browsers that keep their place and ask you when they need you. The monthly subscription starts with a 7-day trial; payment details are required at checkout. Your AI provider charges separately.