> ## Documentation Index
> Fetch the complete documentation index at: https://squad.so/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# September 2026

> Human teammates and more than one squad per account, Slack as a chat channel, the workspace computer for every workspace, and a Model routing card with Claude Opus 5.5 as the new Claude default.

You can now invite people to your workspace and run more than one squad, your agents can join Slack, and every workspace gets the workspace computer. AI Tokens gained a Model routing card, usage meters, and more than one account per provider, and Claude Opus 5.5 leads a large set of new models.

## New

* You can now invite human teammates to your workspace. Open **Settings** > **Team** to invite people by email as a Collaborator, Administrator, or Viewer, and to limit a teammate to some agents. Teammates can read and send chat with the agents you allow, and agents answer each teammate by name. Billing and credits stay with you. Each person's connected accounts stay theirs until that person shares them, and each teammate has their own notification settings. See [Invite teammates to your workspace](/docs/teammates).
* Each scheduled task has a new setting, **Accounts it can use**, once your workspace has teammates. **Shared accounts and my own accounts** lets each run use the shared accounts and the accounts of the person who saved the task. **Only accounts shared with everyone** limits it to the shared accounts. See [Choose the accounts of a scheduled task](/docs/scheduled#choose-the-accounts-of-a-scheduled-task).
* You can now run more than one squad from one account. Open the workspace card at the top of the sidebar and choose **New squad team**. Each squad has its own subscription. See [Run more than one squad](/docs/billing#run-more-than-one-squad).
* Your agents can now join Slack. You pick an agent before you create the Slack app, and the app shows in Slack as that agent. Agents reply in a thread by default in channels, groups, and direct messages, and a toggle on the app card turns threads off. Agents can also send to Slack from the dashboard and from other channels. A **Verify connection** button on each app checks its tokens and permissions and says what to fix. See [Connect Slack](/docs/slack).
* The workspace computer is now on for every workspace. You can watch an agent's screen live, take over and hand back, and teach a task. Each agent has its own browser with its own logins, and the browser opens on a page that names the agent and explains this. See [Watch or take over an agent's computer](/docs/computer).
* **AI Tokens** has a new **Model routing** card. It shows the default model, the fallback models in the order Squad tries them, and every agent that has its own model. You can change the default, reorder, add, or remove fallbacks, and clear an agent's model there. This card is now the only place to edit fallback models, so Agent Configuration no longer has its own fallback list. See [Choose the default model and fallback models](/docs/ai-accounts#choose-the-default-model-and-fallback-models).
* You can now connect more than one Claude Code account. When one account reaches its usage limit, Claude Code runs move to the next account by themselves. You can add the account's email to tell the accounts apart. Gemini, OpenCode, Qwen Cloud, and Workers AI now also take more than one key. Before, a second key replaced the first. Extra Workers AI tokens must come from the same Cloudflare account, and extra Qwen Cloud keys from the same plan. See [Add another account for a provider](/docs/ai-accounts#add-another-account-for-a-provider).
* Account rows on **AI Tokens** now show usage meters for ChatGPT, Claude, Claude Code, GitHub Copilot, OpenRouter, Kimi, and Z.ai accounts. ChatGPT, Claude, Claude Code, and Copilot show how much of each usage limit is used and when it resets. OpenRouter, Kimi, and Z.ai show the credit used or the balance left. Before its first run, a Claude Code account shows the tokens used in the last 5 hours and the last 7 days. See [Read an account row](/docs/ai-accounts#read-an-account-row).
* Gemini now connects with a Google AI Studio API key and brings the four Gemini 3 models. The older Gemini connection is removed. See [Connect your AI subscription](/docs/connect-your-ai).
* You can now connect Cloudflare Workers AI with an API token and your account ID. It runs open models on your own Cloudflare account, among them DeepSeek V4 Pro, GPT-OSS 120B, Llama 4 Scout, Kimi K2.7 Code, GLM-5.3, GLM-5.2, DeepSeek V4 Flash, and Qwen3.8 27B. See [Connect your AI subscription](/docs/connect-your-ai).
* You can now connect OpenCode Zen with one API key. It gives models from several vendors, among them GPT-6.1 Sol, GPT-6 Astra, Claude Opus 5.5, Gemini 3.8 Flash, Grok 4.6, and the free Big Pickle. See [Connect your AI subscription](/docs/connect-your-ai).
* New models are available: Claude Opus 5.5 and Claude Sonnet 5.5 (Anthropic, Claude Code, OpenRouter, OpenCode), GPT-6.1 Sol (ChatGPT, OpenRouter, OpenCode), GPT-6 Sol and GPT-6 Luna (ChatGPT, OpenRouter), Claude Fable 5.1, GPT-6 Astra, DeepSeek V4.1 Flash (OpenRouter, OpenCode), and Grok 4.7. OpenRouter also adds GLM-5.3, GLM-5.3 Flash, Qwen3.8 Max, Qwen3.8 Flash, and DeepSeek V4 Flash. New Grok workspaces still start on Grok 4.6. See [Choose models and manage AI accounts](/docs/ai-accounts).
* A message icon on an agent now opens its conversation in the right-side panel, so you stay on your screen. The message icon in the top bar shows how many conversations have replies you have not read and lists them, and with no unread replies, one click opens your last conversation. A doc that you click in a mission, a profile, or a chat opens in a panel above your screen. See [Top bar and sidebar](/docs/search#top-bar-and-sidebar).
* Chat now has conversation tabs and unread dots. When two or more conversations are open in the side panel, a tab strip lets you switch between them in place, and pointing at a tab shows the agent, the title, and the newest message. A reply that you have not read shows a dot on the agent's sidebar row and on its tab, and the conversation list marks threads with new replies. See [Switch between open conversations](/docs/chat#switch-between-open-conversations).
* You can now pin a conversation to the sidebar. Use the pin in the conversation header or the thread panel, or choose **Pin a conversation** in the command palette. Pins show on all your devices. See [Pin things to the sidebar](/docs/chat#pin-things-to-the-sidebar).
* You can now point at a connected app account in a message. Type **#** in the composer and pick an account, for example Slack for one workspace, so the agent knows which account you mean. Apps that are not connected yet show too. See [Point at missions, docs, and apps](/docs/chat#point-at-missions-docs-and-apps).
* You can now mark a mission as ongoing. Ongoing missions are never flagged as stuck, and **Hide ongoing** on the board hides them. A quiet mission now gets a nudge to its agent first, and you get a ticket only if nothing moves after the nudge. See [Track work on the Missions board](/docs/missions#edit-or-mark-a-mission-as-ongoing).
* The sidebar now has a **Tickets** row with the count of open tickets. In My Tickets, the **Aging** filter shows tickets that have waited on you for more than a day, and you can archive all filtered tickets at once. See [Answer tickets in My Tickets](/docs/tickets).
* **Integrations** has a new **Channels** tab with Telegram, Discord, and Slack. Each card opens that channel's setup screen. See [Talk to your squad from Telegram, Discord, or Slack](/docs/channels).
* Agents now have @handles that you can see and use. The agent profile shows the handle beside the role, and the mention picker in mission comments matches handles. Agents can set a short handle when they create another agent, and @mentions of that handle reach the right agent. See [Rename an agent or change its handle](/docs/agents#rename-an-agent-or-change-its-handle).
* You can now give one agent its own image model, the model that reads screenshots and photos. Set it in the agent's **Image model** setting or in **Model routing** on **AI Tokens**. See [Change an agent's model](/docs/agents#change-an-agents-model).
* You can now connect more than one account of the same MCP server. Each account has its own name, and each server shows one card with an **Add account** control and a row for each account. See [Connect a second account of the same server](/docs/custom-integrations#connect-a-second-account-of-the-same-server).
* Billing now has a **Research tools** card with one switch for all built-in research tools. When the switch is off, the tools stop and use no credits. See [Turn research tools off](/docs/costs#turn-research-tools-off).
* You can now pause an agent's email inbox with **Pause inbox** and turn it back on with **Resume inbox**. Mail that arrives while the inbox is paused waits for the agent. See [Pause an inbox](/docs/email#pause-an-inbox).
* New agent inboxes now get addresses at agents.squad.so. An inbox on the earlier built-in domain can move to agents.squad.so from its domain picker, and the old address then stops receiving. You can also clear the emails of deleted inboxes with **Clear emails from deleted inboxes** in the shared email settings. See [Give an agent an email inbox](/docs/email).
* Squad now emails you about escalations and drafts that wait for your approval, not only the bell. The weekly summary email is now called **Weekly check-in**. See [Know what the weekly check-in contains](/docs/notifications#know-what-the-weekly-check-in-contains).
* You can now pick a keyboard layout on the computer screen. The computer types in the layout you pick, out of 15 common layouts. See [Pick your keyboard layout](/docs/computer#pick-your-keyboard-layout).
* You can now attach files to a mission reply. Use the attach button, drop a file on the composer, or paste it. The files go to the mission's attachments.
* The skill page now shows every file in the skill, not only SKILL.md. Click a file to read it. The details panel stays in view while you scroll. See [Add and manage skills](/docs/skills).
* Documents, missions, and skill pages now have **Hide details**. It hides the details column so the content takes the full width.
* When you cancel your subscription, Squad now asks for a short written reason. Until the end date, **Keep my subscription** undoes the cancellation. See [Undo a cancellation](/docs/billing#undo-a-cancellation).
* The Squad docs now live at squad.so/docs. The in-app Help links there, and your agents answer questions about Squad from the same docs, with a link to the article. See [Get help](/docs/getting-help).

## Improved

* Squad now lives at squad.so, with a new look across the dashboard and the notification emails.
* Home is the start of the dashboard. The top bar logo opens Home, a What's possible preview replaces the credit chart, and the Home composer lets you pick which agent gets your message. The sidebar drops the Home, Runs, What's possible, and Live Feed rows, and Runs moves to a row in Settings. Two new icons at the bottom of the sidebar open Custom integrations and Agent Configuration.
* Claude Opus 5.5 is now the default for new Claude, Claude Code, and OpenRouter workspaces. Claude Opus 5, GPT-5.4, GPT-5.3-Codex, Spark, GLM-5, and GLM-5-Turbo are removed from the model lists. See [Choose models and manage AI accounts](/docs/ai-accounts).
* Your squad now tries every connected account and every fallback model in order before a run fails. Before, an account in cooldown or an unknown error could stop the run early. See [Rate limits and fallback](/docs/agent-stuck#rate-limits-and-fallback).
* The run time limit is no longer a setting. The **Run time limit** card in Agent Configuration says that runs have no time limit, and a limit left from before shows **Remove the limit**. New scheduled tasks get 24 hours, and a task with a shorter limit shows **Raise to 24 hours**. A long Claude Code turn no longer stops after 60 minutes without output. See [Time limits](/docs/agent-stuck#time-limits).
* A paused agent now stays paused. Chat messages to it are refused with a clear message, and @mentions, mission assignments, and emails wait until you resume it. Your lead agent can now pause and resume other agents for you. See [Pause an agent](/docs/agents#pause-an-agent).
* The **Default model** card is now at the top of Agent Configuration, with **Add other models** at the end of its picker. The allowed-models list folds by provider. Model names without a version say that they always run the newest model of their family, and the Claude Code Opus and Sonnet rows say which version that is, for example "latest (now 5.5)". See [Choose the default model and fallback models](/docs/ai-accounts#choose-the-default-model-and-fallback-models).
* Saves in Agent Configuration are now instant. The change applies in the background, and a notice shows until it is live, so you can save again right away. See [Settings reference](/docs/settings#agent-configuration).
* Models now use their full context window, for example 1,000,000 tokens for GPT-5.5, the GPT-5.6 models, GLM-5.3, and MiniMax M3. New Claude models now reach every workspace without a reconnect.
* The lead agent's heartbeat scheduled task now starts switched off in new workspaces. Your lead offers to turn on check-ins, and you can turn them on from the Scheduled screen. See [Turn on heartbeats](/docs/scheduled#turn-on-heartbeats).
* The chat composer is faster and easier to learn. A sent message shows at once, typing @, /, or # opens a list that filters as you type, and the skill list opens at once when you type **/**. The message box grows with long messages, line breaks you type show in chat, the text stays in line with the cursor, and the box shrinks back after you send. You can drop a file anywhere on the conversation or the thread panel. Squad chat and mission comments now use the same composer, so Shift+Enter adds a new line there too. See [Use the composer](/docs/chat#use-the-composer).
* Chat reopens where you left it for each agent, and you can drag the chat panels as wide as you like. Chat times now say when, for example "Yesterday 11:42 AM" or "3 days ago", and pointing at a time shows the full date. See [Chat with your agents](/docs/chat).
* The Runs screen now marks a run that went quiet as **Presumed stopped** in amber instead of a red timeout. If the run reports back, the row corrects itself. When every model in the fallback list fails, the run error now leads with the first failure. See [Read a run's detail](/docs/runs-and-health#read-a-runs-detail).
* Scheduled tasks are easier to manage. The task editor is wider, its instructions box grows with the text, and a custom schedule shows a plain-language preview such as "Runs at 08:00 AM, only on Tuesday and Thursday". On the **Scheduled** screen, the agent name opens that agent's settings, and the error badge shows the error text and opens the failed run on the Runs screen. See [Manage tasks on the Scheduled screen](/docs/scheduled#manage-tasks-on-the-scheduled-tab).
* Setup is more reliable. A slow build now retries on the same computer instead of failing, and **Reset & start fresh** cannot be used while a build runs. After you pay, the page shows "Payment received" until your account is ready. See [Fix setup problems](/docs/setup-troubles).
* A ticket now names the mission it belongs to, and **Discuss with** opens a chat about the ticket in the side panel with starter questions. Links in a ticket's ask are clickable, answered sections fold to one line, and **Provide details** always opens the answer box. See [Answer tickets in My Tickets](/docs/tickets).
* When a Claude Code account signs itself out, the bell rings and AI Tokens shows a banner. In the chat, the raw error becomes one readable line that points to AI Tokens. The run error says to reconnect the account instead of "Context overflow", and the run moves on to the fallback models.
* ChatGPT accounts now renew their sign-in before it expires, so agents keep working after a quiet period. See [Keep ChatGPT accounts renewed](/docs/ai-accounts#keep-chatgpt-accounts-renewed).
* Credits are now quieter. Credit alerts show only in the bell and no longer come by email, the weekly email no longer shows the credit balance, and the navbar no longer shows a credits counter. Billing explains that only the built-in research tools use credits. See [Choose your notifications](/docs/notifications).
* After you cancel, the workspace owner can subscribe again from the lock screen. Checkout opens on the page, and your workspace computer starts again right away. See [Use the lock screen](/docs/billing#use-the-lock-screen).
* Agents now know which chat channels are connected. They post to Slack with the right channel form, and they do not ask you for a channel ID or a bot token.
* Agents now follow your explicit instruction over their skill defaults.
* Integration accounts now show which account they are, including the email of Google accounts, and each account's menu acts on that account. The Integrations screen has one add button instead of several. See [Connect apps](/docs/integrations).
* The page that opens after an MCP sign-in now shows one spinner, closes itself after a clean success, and has a button back to Custom integrations. The key field for a bare server URL is now labeled **Authorization: Bearer**. See [Add an MCP server or an API key](/docs/custom-integrations).
* The agent profile header now has one Message button, icon buttons for the computer and Configure, and a More menu, so the status timeline shows first. Agent avatars are now circles. See [Add, change, and pause agents](/docs/agents).
* Activity rows and bell notifications now show formatted text with clickable links and doc chips. **Mark all read** now clears every notification in the bell at once. See [Use the bell](/docs/notifications#use-the-bell).
* Disconnecting an app or an AI account, removing a secret, and deleting a doc now ask you to confirm first.
* Screens now fit the panel they are in. With a wide chat panel open, layouts drop to one column instead of crushing, and wide tables scroll sideways. When you open a screen from another one, its back button now returns you there and names it.
* Squad works better on a phone. The page no longer zooms in when you tap a text field, and crowded rows, memory cards, and the sidebar fit the screen.
* Settings search now finds words like "secret", "variable", "env", and provider names, and the command palette reaches Agent Tools, Telegram, Discord, Slack, and Public Dashboard. See [Search and keyboard shortcuts](/docs/search).
* Archived missions now have a link back to the board, and you can delete an archived mission (it goes to Trash). See [Archive, delete, and restore missions](/docs/missions#archive-delete-and-restore-missions).
* **Forgot password?** now tells you when no account uses that email, or when the account signs in with Google. See [Fix setup problems](/docs/setup-troubles).
* The sign-in code for device connections now shows first and stays visible while Squad confirms the sign-in.
* The Squad chat room now shows only when you have at least two agents.
* When you invite a Collaborator or an Administrator, the invite dialog offers **Share my connected accounts with everyone in this workspace**. See [Share your accounts when you invite](/docs/teammates#share-your-accounts-when-you-invite).
* In a workspace with teammates, connecting a new app account asks whether to share it with everyone. See [Who can use an account](/docs/integrations#who-can-use-an-account).
* Only the person who connected an app account can now share it or make it private. You and your Administrators keep **Disconnect**, **Make primary**, **Rename**, and **Manage scope**. See [Who can use an account](/docs/integrations#who-can-use-an-account).
* An account that a teammate connected shows **Connected by** and their name. See [Who can use an account](/docs/integrations#who-can-use-an-account).
* **Remove** on a teammate's row now asks you to confirm and says what happens. See [Remove a teammate or revoke an invite](/docs/teammates#remove-a-teammate-or-revoke-an-invite).
* A paused agent now names who paused it, for example "Paused by" and your lead's name. The Activity entry uses the same name. See [Add, change, and pause agents](/docs/agents).
* The workspace card says **Teammate** for a squad that you joined and **Setting up** for a squad that is being built, instead of adding the word plan.
* On busy squads, the **Runs** screen and chat now stay fast. Long run lists draw as you scroll, and a new run or reply no longer reloads the whole list. See [See what your agents did](/docs/runs-and-health).

## Fixed

* Long agent turns no longer get answered twice. Before, a long turn could be marked as failed while it still worked, and a second copy of the agent or the next model answered the same message.
* A run that hangs after repeated rate-limit errors is now stopped, so new messages to that agent no longer wait for hours.
* Schedule presets now save correctly. Before, "Hourly" and "Daily" saved nothing, and "Every 2 hours", "Every 4 hours", and "Every 12 hours" ran every 2, 4, and 12 minutes. Pausing a scheduled task from the Scheduled screen now works too. See [Schedule presets](/docs/scheduled#schedule-presets).
* Five research tool families work again: YouTube search, YouTube comments, LinkedIn profile search, Instagram, and TikTok. See [Research tools reference](/docs/research-tools).
* Busy workspaces with many skills no longer restart again and again, which had stopped runs with "the workspace runtime restarted".
* Claude Code no longer fails with a version error when you pick a newly released model.
* GLM and Grok vision models read images again.
* MCP servers whose tools use a newer JSON Schema version now work.
* Read-only access to an app now includes its read tools, so an agent can, for example, list GitHub pull requests. GitHub code search, issue listing, and issue updates also work at the access levels that include them. See [Choose an access level](/docs/integrations#choose-an-access-level).
* Chat no longer looks frozen after you steer an agent during a tool step. The thread shows that the agent is working until the reply comes.
* Agents now address you by your name, not by the workspace name.
* An @mention typed in the first message of a new conversation now creates the group chat.
* Files you attach to a message no longer attach again to your next messages. Long ticket names in a chat no longer wrap onto a second line.
* **Export as PDF** on a doc now shows the formatted document, not the raw text. Exported documents now say Squad in the footer. See [Export a doc](/docs/squad-docs#export-a-doc).
* Runs now shows all research and tool events in its window, and the Integrations, Research, and Scheduled-runs filters hide the rows they name. See [Find a run](/docs/runs-and-health#find-a-run).
* The calendar week view stays readable when many tasks run a few minutes apart. See [See your week on the calendar](/docs/scheduled#see-your-week-on-the-calendar).
* The Docs badge now counts docs that changed since you read them. See [Read-state badges](/docs/squad-docs#read-state-badges).
* **Settings** > **Integrations** now counts your connected apps. Before, it always showed "1 connected".
* Pasted keys and tokens that wrapped onto several lines in a terminal are now accepted.
* When the checkout for **New squad team** cannot open, a message now shows under the menu item. Before, nothing happened. See [Run more than one squad](/docs/billing#run-more-than-one-squad).
* Agent browsers on some workspace computers could not start, so every page that an agent opened failed. They start again.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.