# Changelog

What is new, improved and fixed in AI Workforce One HQ, the hq CLI and the VS Code extension, newest first.



What changed in HQ, the `hq` CLI and the AI Workforce One HQ extension for VS Code, newest first.
Dates are in US Eastern time.

* **VS Code extension:** every version's notes are on the
  [Visual Studio Marketplace](https://marketplace.visualstudio.com/items/aiworkforceone.awo-hq/changelog)
  and on [Open VSX](https://open-vsx.org/extension/aiworkforceone/awo-hq/changes). To update, see
  [Install the extension](/docs/vscode/install/).
* **hq CLI:** each release is listed below on the day it came out. To update, see
  [Install the CLI](/docs/cli/install/).
* **HQ API:** changes to the API are listed in the [API changelog](/docs/api/changelog/).

## October 29, 2026 [#october-29-2026]

### Fixed [#fixed]

* The browser view no longer pauses after 30 minutes while your AI Employee is using the browser.
  If it did pause, it now comes back by itself as soon as your AI Employee uses the browser again,
  without reloading the page.

## October 25, 2026 [#october-25-2026]

### Fixed [#fixed-1]

* A Cloud Computer that stayed on for more than about a week could stop opening new sessions and
  the Files panel, showing "Couldn't start the Cloud Computer" while its running sessions kept
  working. Its clock now stays correct for as long as it runs. A Cloud Computer picks this up the
  next time it restarts.

## October 24, 2026 [#october-24-2026]

### Fixed [#fixed-2]

* A scheduled run no longer restarts a session that has its terminal open. The run is skipped, you
  are notified, and the terminal and the work in it are left as they are.
* When a job uses up a Cloud Computer's memory, that job is now stopped instead of the Cloud
  Computer's own services.
* Opening a session just as its Cloud Computer was going to sleep no longer leaves the terminal
  disconnected for a minute or more after the Cloud Computer is back. The session reconnects within
  seconds of waking.
* **hq CLI 0.3.2 and VS Code extension 0.4.2:** waking a Cloud Computer of 80 GB and up from `hq up`,
  `hq attach`, `hq forward` or VS Code no longer gives up after 3 minutes. They now wait up to 4
  minutes, as long as these sizes can take to wake. They also stop with HQ's message, instead of
  retrying, when a connection already has as many sessions open as HQ allows.
* Saving a team as a template and connecting your ChatGPT account no longer stop after 2 minutes
  while a Cloud Computer of 80 GB and up is still waking. They wait up to 4 minutes and carry on by
  themselves once it is back.
* Saving a team as a template and connecting your ChatGPT account also wait long enough for a large
  Cloud Computer set up for your organization by contract, and for one moved to a smaller size that
  has not restarted yet and still runs at its larger size.
* Chat now lists the tools Claude loaded when you open a "Loaded tools" card.

## October 23, 2026 [#october-23-2026]

### New [#new]

* Larger Cloud Computers, up to 128 GB of memory or 128 processor cores, are now available on
  request. Contact support and we can turn them on for your organization.

### Improved [#improved]

* When your Cloud Computer is on the largest size and runs short of memory or processor, HQ now tells
  you that bigger sizes are available and links to support.
* Every Cloud Computer size runs more sessions at once: 5 on Standard, 12 on Boost, 24 on Boost Pro
  and 40 on Boost Max, and 40 to 68 on the larger sizes.
* When a team's Cloud Computer is too busy for another session (memory nearly full, or the processor
  working at full capacity), HQ now says so in plain words and suggests stopping a session or moving
  the team to a bigger size, instead of slowing every session down.

### Fixed [#fixed-3]

* Run now on a schedule whose Cloud Computer was asleep now wakes the Cloud Computer and runs the
  prompt, instead of failing after five and a half minutes.
* A scheduled run into a brand-new session, or the first run after `/clear`, is no longer reported
  as not delivered when Claude did reply.
* A scheduled run no longer stops when Claude asks "Resume this conversation?" on a session that has
  been idle for a while. HQ now resumes the same conversation and the run continues, and anything you
  had typed but not sent stays where it was.
* Waking a large Cloud Computer (80 GB and up) no longer stops at 2 minutes with a Retry button. The
  session page keeps waiting up to 4 minutes and connects by itself when the Cloud Computer is back,
  and anything you typed while it woke is sent then.

## October 22, 2026 [#october-22-2026]

### Improved [#improved-1]

* Schedule run details now say clearly when the target session was deleted, when a run is waiting
  for the input box, and when a run failed.
* Reassigning a session can now move its schedules to the new owner, and a schedule stopped by a
  reassign can be moved with one click.
* On a phone, the Profile and People tabs now scroll to show the tab you are on.
* Session pages and the terminal open faster on a first visit in Chrome, Edge and other Chromium
  browsers.
* The app loads faster on a first visit, especially on slower connections.
* Chat now shows which tools Claude loaded instead of a generic tool card.

### Fixed [#fixed-4]

* On a narrow screen, the Browser View address bar now shortens a long site name from its start,
  and shortens `http://` and a port number before the site name, so the end of the name stays
  visible, and a site that is not secure is marked.
* You can now invite someone again after their earlier invitation expired.
* When your Cloud Computer's processor stays fully busy for more than about ten minutes, or it runs
  short of memory, HQ now suggests a bigger size as intended. Before, the suggestion often did not
  appear.
* Scheduled runs now continue a long, idle conversation automatically instead of stopping at
  Claude's resume question.
* When Claude asks whether to resume a long, idle conversation, Chat now shows the question with a
  **Resume** button for the person who types in the session. To start a new conversation instead,
  use the terminal.
* Only the person a schedule runs as can now change its prompt or the session it runs in, also on a
  schedule that starts a new session each run.
* Reloading a session page while Claude is working no longer leaves the terminal blank with
  "Couldn't restore the previous screen". The screen comes back with any secret hidden, in plain
  text until Claude redraws it.
* The browser view now stays open when you reload the page, and it keeps reconnecting by itself
  while HQ is being updated instead of closing.
* Sign-in tokens printed in a Codex session's terminal are now hidden reliably, even when they
  arrive in pieces.
* Upgrading your plan, buying add-ons and starting a paid plan early on a higher plan now work with
  a card your bank asks you to confirm. HQ opens Stripe's secure payment page for the confirmation
  instead of saying the card was declined, and nothing changes until the payment goes through.
* A team secret is now also hidden in the terminal when Claude wraps a long value onto the next
  line, including when you reload or reopen the session.
* Secrets are now also hidden in colored terminal output (for example from grep, bat or jq) and when
  a long sign-in token wraps over several lines on a narrow screen.
* API keys and sign-in tokens are now also hidden when they wrap onto several lines in the terminal.
* A saved secret is now also hidden in the terminal when coloured output (for example from grep)
  highlights part of it.
* A secret you change or delete now stays hidden in older session history and transcripts, however
  many times it changed. Deleting a secret always works.
* A downloaded session history now keeps hiding a secret that was changed many times while the
  session was running.
* A paused or closed account now shows the same clear notice on every page, and a closed account is
  no longer told its data was already erased.
* On a closed account, People → Roles and a member's People page now show the account notice instead
  of a loading error.
* Reactivating your account with a card that needs your bank's confirmation (3D Secure) now takes
  you to a secure Stripe page to confirm, instead of saying the card was declined.
* New Cloud Computers no longer show an "Update available" notice for Claude Code right after they
  start.
* Long feedback typed into a plan or question card is no longer cut short.
* An AI Team role no longer lists permissions only an organization role can hold, such as deleting
  AI Teams, so **All** no longer stops the role from saving.
* The role editor's "selected" count now counts only the permissions it shows.
* The action filter in the audit log and the Roles screens no longer list actions and permissions
  for features your organization does not have.
* Removing someone's access to a shared session now stops their view within 30 seconds.
* A session whose start was interrupted no longer stays on Starting; it now shows as failed with a
  retry.
* A Cloud Computer no longer goes to sleep while a program on it is still busy using the processor.
* Stopping and quickly starting a team no longer leaves its earlier sessions looking like they are
  still running.
* On a tablet, reloading a session with Browser View open no longer shows a garbled terminal for a
  few seconds.
* The Claude sign-in panel now closes by itself after you sign in, in every permission mode.

## October 21, 2026 [#october-21-2026]

### Improved [#improved-2]

* The details of a scheduled run now read in plain words. They say what happened and what you can
  do about it, for example when a message is waiting for approval or the session is not signed in.

### Fixed [#fixed-5]

* Restarting a session on a paused team no longer stops it.
* A message from another session that Claude is holding now gets **Deliver** and **Deny** in
  Chat on every device, also when it is long or contains quotes or brackets. Before, Chat showed
  only the live screen for those. Chat shows the part of the message Claude previews, says how much
  of it that is, and **Deliver** sends the whole message.
* In Chat, **Open Terminal**, **Show details**, **Try again** and **Dismiss** are now easy to tap on
  a phone and a tablet.
* After `/exit`, the kept screen no longer shows Claude's command menu or a "claude --resume" line
  that can't be used in HQ.
* When you turn down a plan, Chat now says "You didn't approve the plan" and marks the plan "not
  approved", instead of "Claude was interrupted" and "it did not work".
* When you stop Claude or say no to a permission, Chat now says you stopped it and marks the step
  "stopped by you", instead of "Claude was interrupted" and "it did not work".
* On a tablet, **Restart** and the close button in a session's list of notices are now easy to tap,
  as on a phone.
* If a Cloud Computer can't load its latest settings when it starts, Docker now stays off unless HQ
  last allowed it, and the Docker card says why. Sessions keep working, and **Restart Cloud
  Computer** tries again. See [Docker](/docs/cli/docker/).
* When you turn down a plan with feedback in **Tell Claude what to change**, Chat now marks the plan
  "not approved" instead of "it did not work".
* On a tablet, the **Send** button in Chat and the notice buttons at the top of the page are now
  easy to tap.
* When you stop Claude from writing or changing a file, Chat now says "Did not write" or "Did not
  change" that file, instead of saying it was written.

## October 20, 2026 [#october-20-2026]

### Improved [#improved-3]

* Cloud Computers now come with ffmpeg and unzip preinstalled, so video and archive tools work after
  every restart.
* When the AI Employee ends a session itself, for example after `/exit`, the terminal keeps its last
  output on screen with **Start** ready, instead of switching to the stopped screen.
* A long build in a terminal session now keeps the Cloud Computer awake while it is using the
  processor, even when it prints nothing for a while. See
  [When a Cloud Computer sleeps](/docs/ai-teams/sleep/).
* A schedule prompt can no longer contain hidden control characters (tabs and line breaks are
  fine), so a scheduled run always sends exactly the text you see.
* You can now tell Claude what to change about a plan right in Chat. Under **Approve plan**, type
  your changes in **Tell Claude what to change** and press **Send**. Claude keeps planning with
  your feedback and shows you a new plan.
* Scheduled runs show Success only when Claude actually received the prompt and started on it.
  Otherwise the run says what happened instead, for example that Claude did not take the prompt, the
  session was unavailable, a message was waiting for approval, or the usage limit was reached.

### Fixed [#fixed-6]

* Chat and schedules no longer refuse to send after Claude replies with a table.
* Answering a question with your own text in Chat no longer sends an earlier answer by mistake, and
  typing a new answer over an old one now replaces it completely.
* When the session's terminal is narrow, as on a phone, you can now answer a question in Chat with
  your own longer text. Before, Chat could not read the question once your answer wrapped.
* On a phone or a tablet, a message from another session that Claude is holding now shows in Chat
  with its **Deliver** and **Deny** buttons, like on a computer. Before, Chat showed only the live
  screen there. On a computer the buttons also appear sooner.
* In Chat, a "Thinking…" line with nothing to open no longer stays behind once Claude's reply has
  moved on, so finished turns no longer pile up "Switch to Terminal to see the reasoning" lines.
* When Claude reaches a usage limit, Chat shows the reset time in your own time zone instead of the
  Cloud Computer's (UTC).
* When a Claude request fails (a usage limit, an expired login, a busy service), Chat now shows it as
  a highlighted notice in the conversation instead of a faint line.
* On phones, the Create session button on the new session page can be reached again.

## October 19, 2026 [#october-19-2026]

### New [#new-1]

* The HQ API is here. Your own code, scripts and CI jobs can now read and drive your organization
  at `https://api-hq.aiworkforceone.com/v1`: your AI teams, their sessions, schedules and Cloud
  Computers, and more, with signed webhooks when things happen. API keys are on every plan. During
  a trial the API can only read and you can hold one key; choose a plan to make changes. Create a
  key in **Profile**, **API keys**, or (owners and admins) a service key for your organization in
  **Settings**, **Developers**. Start with the [API overview](/docs/api/) and the
  [quickstart](/docs/api/quickstart/).
* OAuth apps: let your own product act for people in HQ, in the organization they pick, after they
  approve it. Registering an app needs the Growth plan or higher. See
  [OAuth apps](/docs/api/oauth/).
* When one of your Claude sessions holds a message from another session (their permission modes
  differ), Chat says a message is waiting and shows it with **Deliver** and **Deny** buttons. The
  message is shown whole, as plain text. A message Chat cannot show whole and unchanged (a long
  one, or one with a session address in it) is answered in the terminal instead, where you can read
  all of it. Nothing is delivered until you choose, and a message nobody answers expires after a
  few minutes (the sender is told it was declined).

### Improved [#improved-4]

* AI Employees now know you can watch their browser in HQ. When the browser is on for a team, a new
  session tells your AI Employee that you can follow along in **Open the browser**, to say what it
  is doing, to ask you to do any sign-in step yourself, and never to ask for a password or a code in
  the chat. See [Browser View](/docs/sessions/browser-view/).
* AI Employees now know where to install tools on a Cloud Computer so they are still there after a
  restart, an update or a wake from sleep, and that the machine has no `sudo` or `apt`. This applies
  to new sessions.
* In Chat, a note from another Claude session names it by its name only, never by its internal
  address, and opening a note no longer repeats its first line when the line already fits. Several
  "Thinking…" lines in a row now show as one line that opens to all of it, and a helper's card says
  what it was asked in plain words.
* When Claude asks to fetch a web page, Chat shows the address and **Allow** / **Don't allow**
  buttons instead of the terminal screen.
* When Claude has a plan ready, Chat shows an **Approve plan** button. Approving keeps Claude asking
  before each edit; it never turns on auto mode. To change the plan, answer it in the terminal.
* When Claude cannot answer in Chat, it now says why and what to do: Claude is busy, you reached a
  usage limit (and when it resets), your Claude login expired, the conversation is too long, or the
  connection dropped. A job Claude ran in the background now shows a short line in Chat when it
  finishes.

### Fixed [#fixed-7]

* Tools installed in the wrong folder on a Cloud Computer now fail right away with a clear message
  instead of disappearing at the next restart. Install your own tools in your home folder, which is
  kept: see [Storage](/docs/ai-teams/storage/#what-is-kept-when-the-cloud-computer-restarts).
* Claude's questions and permission prompts show as buttons in Chat again.
* Messages from your other Claude sessions now show as a short note in Chat instead of raw text.
* Stopping a turn, from Chat or the API, no longer declines a message from another session that
  Claude is holding for you. While one waits, Stop does nothing and you answer the message first.
* In Chat, your message now always shows above Claude's reply to it. Some replies showed above the
  message they answered.
* HQ now answers a malformed request with a clear error instead of a server error.
* On iPhone and iPad, the Files panel keeps keyboard focus inside it while open.
* On phones and tablets, screen readers again read out notices and status updates during a session,
  also while the Files panel is open.
* On small phones, the schedule frequency choices (Once, Daily, Weekdays and so on) are no longer cut
  off: a choice that does not fit moves to a second row.
* On an iPad held upright, the note next to **Change plan** on the Billing page now wraps instead of
  running off the card.
* When control of the browser returns to your AI Employee after 30 minutes, the browser view now says
  so, instead of only "Control returned to the AI Employee." The same goes for the other reasons
  control ends, such as 5 minutes without activity.
* On a team that keeps each member's work separate, the browser of a session on your own Claude login
  no longer says other members of the team can reach it. That browser runs in your own part of the
  Cloud Computer.
* Saved sign-ins: **Last used** now changes only when a sign-in is actually used in a browser
  session.
* A schedule no longer types into a session that is waiting for you to deliver or deny a message
  from another Claude session. The run now shows as **Not run** with the reason, and you're
  notified once until the message is answered. Before, the scheduled prompt was lost and the waiting
  message was declined. Chat and the API hold your message the same way, and Chat says where to
  answer. See [Schedules](/docs/schedules/#when-a-session-is-waiting-on-a-message).

## October 18, 2026 [#october-18-2026]

### New [#new-2]

* Take control of your AI Employee's browser: select **Take control** under the live view to click,
  type and scroll in the page yourself, then **Hand back** and your AI Employee carries on. Only the
  session's owner can take control of a session on their own Claude login. See
  [Browser View](/docs/sessions/browser-view/).
* **Needs you:** when your AI Employee needs you to sign in, enter a code, solve a CAPTCHA or
  confirm something in the browser, it asks at the top of the session and waits. Select
  **Open the browser**, do the step and hand it back, so your password never goes through the chat.
* When a page in the browser asks for a file, you can choose files from your session's files and
  attach them, while you have control. Secret files and keys are never offered.
* Saved browser sign-ins: sign in to a site once and your AI Employee's browser stays signed in
  across sessions and restarts. Sign-ins are encrypted and kept per person and per team, and you can
  sign out of one site or clear everything in the team's **Settings**.

### Fixed [#fixed-8]

* Stopping a Cloud Computer with Docker turned on is faster: stopping it, letting it sleep or deleting
  its team no longer waits about a minute and a half for Docker to shut down. Your containers are
  still given time to stop cleanly.
* On a phone, when a session stops unexpectedly while you have it open, **Start** now shows at the top
  of the page, so you can start it again without opening the status menu.
* Screen readers no longer reach the closed file list on phones and tablets.

## October 17, 2026 [#october-17-2026]

### Improved [#improved-5]

* The sidebar now says how many sessions are running in a plain sentence, kept apart from your storage.
* HQ works better on phones and tablets: forms no longer zoom in on iPhone, the Secrets and Variables
  tables fit small screens, and buttons are easier to tap.
* On phones and tablets the terminal now scrolls with momentum, like the rest of your phone.
* The top of a session is now one slim bar on every screen. It shows the session's name, its team and
  one status you can select for **Restart**, **Stop session** and the Cloud Computer's memory and
  processor. On a phone, **Chat** and **Terminal** are one tap apart and everything else is under
  **More**, so the conversation gets more of the screen. The message box and the terminal keys are
  slimmer on a phone too. On a tablet, the session now fills the whole screen and the file list opens
  from **More**.
* A session that ran out of memory now says **Out of memory** at the top, also after you reload the
  page.

### Fixed [#fixed-9]

* When a session's AI Employee crashes while you have it open, its terminal and its last output now
  stay on screen, with the session marked as stopped unexpectedly and **Start** ready, instead of the
  screen being cleared.
* A session whose agent stops during start now says it could not start, instead of asking you to
  update the Cloud Computer.
* When a session that is not signed in to Claude crashes, the message saying it stopped unexpectedly
  now shows instead of staying hidden behind the Claude sign-in panel. Press **Start** and the sign-in
  panel comes back.
* On phones, opening the keyboard in a session no longer hides the screen, and the message box and
  **Send** stay in view, in both Chat and the terminal.
* On a touch screen with no trackpad or mouse attached, pressing Return in the Chat message box now
  adds a new line. Tap **Send** to send, or press Cmd+Enter (Ctrl+Enter) on a keyboard.
* On iPhone, tapping the terminal in a session no longer zooms the page in.
* When a message can't be sent yet because Claude is asking something on screen, Chat now says so
  just above the message box and keeps your message, so you can answer first.
* On phones, the stopped and crashed session panels now fit the screen, and a stopped session shows
  **Start** without a greyed out **Stop** beside it.
* On iPad in portrait, an empty file panel no longer takes space below the session.
* The session's name now stays readable at the top of the page on phones and iPad.
* On small phones, **Jump to latest** no longer covers the Claude sign-in steps.
* The dashboard greeting now says "No sessions running" with a grey dot when nothing is running,
  matching the sidebar, instead of "0 sessions running" beside the green running dot.
* On slower phones, flicking repeatedly through the terminal speeds up again.

## October 16, 2026 [#october-16-2026]

### Improved [#improved-6]

* Professional now includes unlimited team members. There are no seats to buy on Professional, and
  if you move up to Professional with seats from Growth, they are removed from your bill. See
  [Billing and plans](/docs/billing/).
* HQ now notices when Docker on an AI Team stops responding, or can no longer start containers, and
  offers a restart. See [Docker in a team](/docs/cli/docker/).
* The **Data export** page now explains exactly what is kept when an organization is erased: each
  person's own sign-in account, and a recovery copy for 30 days.
* The audit log now shows when a session ended on its own, for example when its AI Employee
  finished and exited, so a stopped session is never left unexplained.
* If your role doesn't allow connecting from a computer on any AI Team, `hq attach` and the VS Code
  extension now tell you so right away, instead of connecting and then dropping the connection.

### Fixed [#fixed-10]

* Switching a session between Claude and the terminal no longer stops it if the switch can't happen
  right away, for example while the AI Team's Cloud Computer is still finishing an update. The
  session keeps running as it was, and you can switch again once the update is done.
* Changing someone's role now also ends any remote terminal they already had open (`hq attach` or
  the VS Code extension) if the new role no longer allows connecting from a computer. If it no
  longer allows it on any AI Team, their computer is unlinked from the organization and needs to be
  linked again once the role gives the access back.
* When a role change takes away connecting from a computer, the VS Code extension now stops
  receiving file updates right away, instead of up to 30 seconds later. When a remote connection
  ends because its access was removed, the audit log now says so (reason "access lost").
* Turning off a permission in a custom role that another permission still needs now tells you which
  one needs it, instead of saving without the change.
* New AI Teams created at the same moment in a new account could fail to start; they now start
  reliably.
* Your organization's storage figure now lines up across the sidebar, **Billing**, and each AI
  Team's page and settings. Until your AI Teams' storage has been measured, they all say "Not
  measured yet" instead of showing 0 B.
* A session whose AI Employee exits on its own now shows as stopped right away, on its page, in the
  **Sessions** list and in the sidebar's count of running sessions, and no longer takes up a running
  session slot until someone opens it again. If the AI Employee crashes instead, or is shut down
  because it ran out of memory, the session now shows as an error right away, so you can see it
  needs attention and start it again. The message
  shown when a session stops unexpectedly no longer says it restarts by itself: it tells you to
  start the session again.
* A Boost bought while the Cloud Computer was updating now applies as soon as the update finishes.
* After a Cloud Computer update or restart, the activity log now shows each session it brought back
  once, instead of twice a second apart.
* Paging through the audit log and schedule run history, in HQ and in the public API, no longer
  skips entries recorded at the same moment.
* In the public API, a date filter (`created[gte]`, `created[lte]`) with an invalid date now returns a
  clear `400` error instead of a server error.
* When a running session's AI Employee crashes or runs out of memory, its page now says the session
  stopped unexpectedly (or ran out of memory) and offers **Start**, instead of saying the Cloud
  Computer couldn't start.
* On a trial, the Boost dialog now says up front that a Boost starts with your paid plan, with a link
  to start it, instead of promising the Boost and then refusing it.
* The order summary on **Billing** and in the Boost dialog no longer lists a $0 seats line on
  Professional, which includes unlimited team members.
* On **Billing**, an invoice for a plan or add-on change is now labeled by the change, for example
  "Plan change to Professional" or "Add-on change", instead of by its first line item.

## October 15, 2026 [#october-15-2026]

### New [#new-3]

* **HQ API reference.** The reference for the HQ API is now public: every endpoint with its
  parameters, example requests in cURL, JavaScript and Python, and example responses, plus guides
  for authentication, versioning, errors and SDKs. See the [API reference](/docs/api/reference/).
* **Guides for HQ.** Step-by-step guides with screenshots for everything you do in HQ: a 5-minute
  quickstart, AI Teams and Boost, sessions, Preview, Browser View, schedules, secrets, people,
  billing and security. Start with [What is HQ](/docs/get-started/). The docs are now grouped
  under **Guides**, **Developers** and **Changelog**, and every page can be opened in Claude or
  ChatGPT.
* **Help in HQ.** A Help menu and Learn more links take you to the right guide from the main pages.
  Open **Help** at the bottom of the sidebar for the docs, what's new and support.
* **Custom roles.** Create roles with exactly the permissions people need, for the whole
  organization or for one AI Team. Start from a template or a system role, preview HQ the way a role
  sees it, and check why someone can or can't do something. On Professional and Enterprise plans,
  once a trial has converted. See [Roles and permissions](/docs/people/roles/).

### Improved [#improved-7]

* Roles moved from Settings to People: open **People**, then the **Roles** tab. A person's role on
  **People** links to that role. Old links still work.
* **hq CLI 0.3.1:** if you can no longer type in a session while `hq attach` is open, it keeps
  showing the session instead of exiting on your next key. What you type is not sent, and it tells
  you that you can only watch.
* **hq CLI 0.3.1 and VS Code extension 0.4.1:** when your access to a session ends, they say "Your
  access to this session ended." instead of "Session not found".

### Fixed [#fixed-11]

* Removing someone from an AI Team now ends their access to that team's sessions right away,
  including the sessions they started themselves.
* Changing a stopped AI Team's Cloud Computer size now applies in one start, a sleeping team keeps its sessions while it changes size, and
  Restart now to apply always tells you what it did. See
  [Cloud Computer sizes and Boost](/docs/ai-teams/cloud-computers/).
* HQ now respects each AI Team's session visibility setting everywhere. On a team set to
  **People see only their own sessions**, the dashboard, the team's session list, the session
  counts on AI Team cards and each session's own page show only the sessions you can see on the
  **Sessions** page.
* Team owners can now open every session their team's session list shows them, as
  [Who can see sessions](/docs/ai-teams/settings/) describes.

## October 14, 2026 [#october-14-2026]

### New [#new-4]

* **Start, stop and restart sessions from your terminal.** hq CLI 0.3.0 adds `hq sessions`,
  `hq start`, `hq stop` and `hq restart`, with the same rules as HQ: the session owner or a team
  owner can run a session. `hq stop` and `hq restart` ask first while Claude or Codex is answering.
  See [Sessions](/docs/cli/sessions/).
* **Start, Stop and Restart in VS Code.** Extension 0.4.0 adds them to each session in the Teams
  view, shown only on sessions you may run.
* **Disk warnings.** When an AI Team's disk is 85%, 90% or 95% full, the session header and the AI
  Team's card say so, measured against how far the disk can still grow, with what you can do to make
  room.
* The Audit log shows when a session was started, stopped or restarted from the `hq` CLI, and from
  which computer.
* **Sign out** on the organization chooser, for people who belong to more than one organization.

### Improved [#improved-8]

* Waking an AI Team says how long it usually takes for its size: 15 to 30 seconds up to 16 GB, and
  30 to 45 seconds from 32 GB. In HQ, the CLI and VS Code.
* **Turn off browser** takes effect right away, also in sessions that are already running, and the
  browser stays off after the AI Team Cloud Computer sleeps, until you enable it again.
* A session that is still starting shows as starting in the CLI and VS Code, instead of looking like
  it is already running.
* The CLI says how long to wait after too many starts in a minute, for example "Too many requests.
  Try again in 34 seconds."
* VS Code says "AI Team Cloud Computer" where it said "team machine", and an update or size change
  says it usually takes about 5 to 6 minutes instead of counting down.

### Fixed [#fixed-12]

* The file explorer waits while a session is starting instead of reading a Cloud Computer that is
  still booting, stops checking after you stop the session, and no longer tries to load files while
  the AI Team Cloud Computer is off.

## October 13, 2026 [#october-13-2026]

### New [#new-5]

* **Pooled storage.** Your organization's storage becomes one pool that its AI Teams share: a
  team's disk grows from the pool as it fills, and owners and admins can set a cap or a reserve for
  each AI Team. It is reaching organizations over the coming weeks. See
  [Storage for AI Teams](/docs/ai-teams/storage/).

### Improved [#improved-9]

* `/tmp` and the files of the browser your AI Employees use, including downloads, now live on the
  AI Team's disk instead of in a small amount of memory, so large temporary files and downloads have
  room. Each AI Team Cloud Computer picks this up at its next update.
* Opening a session on a sleeping AI Team waits up to 2 minutes for it to wake and connects by
  itself, and wakes are noticed sooner.
* A revealed secret value is shown in full, with **Copy**, on organization and team secrets.
* Start and Stop on a session show only to people who may use them.
* If you belong to more than one organization and sign in from a share link, you return to the
  link after choosing an organization.

### Fixed [#fixed-13]

* Selected text in the file editor is visible and readable.
* Coming back to a browser tab with a live terminal no longer clears and redraws it.
* A key typed after clicking into a session's history goes back to the live view and stays there.
* A terminal could resize back and forth and show Claude Code's output twice. It no longer does.
* Audit log CSV exports are safe to open in spreadsheet apps.

## October 12, 2026 [#october-12-2026]

### Fixed [#fixed-14]

* Opening a template share link directly in a new tab works the first time.

## October 11, 2026 [#october-11-2026]

### New [#new-6]

* **Turn off browser.** After the browser is enabled for an AI Team, **Turn off browser** removes it
  again.
* **Open** on every row under **Other ports** in the Preview port picker.
* Before an AI Team is created from a shared template, HQ lists the hooks, connectors and commands
  the template can run on the team's Cloud Computer, and you confirm each one.
* Deleting an AI Team says what happens to its files: they are kept for 30 days in case you change
  your mind, then erased. Owners get an email when a deleted team's files are erased.

### Improved [#improved-10]

* Session notices sit in the header controls, so the terminal no longer changes size when one
  appears.
* HQ waits for Claude to finish printing before it resizes the terminal, so a resize no longer
  garbles the output.
* AI Team actions show one clear pending state while they run.
* On phones, **Restart** is under **More**, and the session header fits small screens.
* When every session slot is busy, a scheduled run waits up to 10 minutes for a free one. A run more
  than 15 minutes late is skipped and the schedule moves on to its next time.
* An HQ tab left open with nobody using it no longer wakes a sleeping AI Team.
* A plain loading screen shows while an AI Team Cloud Computer wakes or reconnects.

### Fixed [#fixed-15]

* Only an owner can remove another owner from an organization.
* Deleting an AI Team with schedule run history works again.
* A paid Boost is always applied, and assigning a spare Boost you already own works directly.
* An upgrade is reported as done only once the size really changed, and the time left never goes up.
* What you type while an AI Team Cloud Computer wakes is no longer lost.
* A schedule created paused shows no next run, and an overdue run shows "due now".
* Browser View shows the page once it settles, and comes back after the Cloud Computer wakes.
* Text people type, such as names, always appears as plain text in HQ emails.

## October 10, 2026 [#october-10-2026]

### Improved [#improved-11]

* **hq CLI 0.2.1:** `hq login --no-browser` prints the sign-in link and code instead of opening a
  browser (automatic on Linux without a display). After signing in, `hq` lists what to run next.
  `hq attach` no longer prints "connection lost" once an hour, redraws at your size when the session
  is also open elsewhere, and says when the session stops.

## October 9, 2026 [#october-9-2026]

### New [#new-7]

* Deleting an AI Team releases its purchased Boost automatically.

### Improved [#improved-12]

* Progress shows from the moment you confirm an AI Team Cloud Computer upgrade.
* An AI Team whose Claude session is signed out can now sleep when idle.
* The Claude Code update notice clears as soon as the restarted session is running.

### Fixed [#fixed-16]

* Claude's own sign-in callback no longer shows up as an app in Preview.
* Members no longer see an empty Organization heading in the sidebar.
* Coming back to a tab reconnects, wakes the Cloud Computer and keeps what you typed.

## October 8, 2026 [#october-8-2026]

### New [#new-8]

* **Session history.** Scroll back to the start of a session, shown exactly as it looked live.
* The session header shows processor use next to memory, with a warning when the Cloud Computer is
  overloaded and a way to upgrade it.
* Your profile has tabs for **Security** and for your personal secrets.

### Improved [#improved-13]

* HQ calls the team's computer the "AI Team Cloud Computer", and agents "AI Employees".
* A processor or memory warning ends as soon as it is no longer true.

## October 7, 2026 [#october-7-2026]

### New [#new-9]

* **My secrets and variables.** Set personal secrets and variables, or import a `.env` file. They
  apply only to sessions you own that sign in with your own account, and take the place of a team or
  organization value with the same name.
* The Preview port picker groups many listening ports so they are easy to find.

### Fixed [#fixed-17]

* A secret's earlier value stays masked after the secret is changed or removed.
* Session history keeps your place while it loads and keeps every line after a resize.
* Signing in to Claude for the first time from the Chat tab is calmer and clearer.

## October 6, 2026 [#october-6-2026]

### Improved [#improved-14]

* The file explorer loads one folder at a time, so large AI Teams open quickly.
* Team owners can run their own team: rename it, update and restart its Cloud Computer, and manage
  who has access. Deleting sessions is for organization owners and admins.

### Fixed [#fixed-18]

* People who are not admins no longer see email addresses in the reassign and share lists.

## October 5, 2026 [#october-5-2026]

### Improved [#improved-15]

* A stopped session says it is stopped and shows **Start**. Opening it never starts it by itself.
* A session open in two windows no longer garbles: the window you are using sets the size and the
  others match it.
* **VS Code extension 0.2.3:** the hourly reconnect is invisible, with no keystroke lost.

## October 2, 2026 [#october-2-2026]

### New [#new-10]

* **VS Code extension 0.2.1:** more than one HQ account on the same computer, one per window.
* **VS Code extension 0.2.2:** paste or drop images into session terminals.
* Syntax highlighting in the file editor for more than 40 languages.
* An AI Team's **Variables** tab shows the organization variables the team receives.
* Secrets and variables show one table per tab with a level filter, and each AI Team can import a
  `.env` file.

### Improved [#improved-16]

* Docker storage on an AI Team Cloud Computer grows by itself as the team's disk grows.
* An HQ tab counts as someone being there only while you are using it, so an idle AI Team can sleep.
* AI Teams with idle MCP servers, such as Playwright, can now sleep.

## October 1, 2026 [#october-1-2026]

### New [#new-11]

* **Restart Cloud Computer.** Owners and admins can restart an AI Team Cloud Computer at any time,
  and running sessions come back afterwards.
* Preview finds ports from what is actually listening, names the Docker container behind each one,
  and lists services that are not websites, such as Postgres or Redis, under **Other ports**.
* Changing an AI Team Cloud Computer's size brings your sessions back afterwards.
* A first-run Claude sign-in card in the Chat tab.
* Search in the organization picker when you belong to more than 6 organizations.

### Improved [#improved-17]

* The session header names the AI Team and links to it.
* The memory indicator shows the Cloud Computer's real memory.
* An organization that requires two-factor authentication requires it everywhere in HQ until the
  second step is done, and first-time members get a setup page.

### Fixed [#fixed-19]

* Session header controls work again on phones.
* Dialog fields keep your cursor when the page refreshes in the background.
