New
- The Squad chat is now a live room. A message reaches every active agent at once, and an @mention narrows it to the agents you name. Replies show while the agent writes them. See Use the Squad chat.
- You can start group chats with any agents of your squad. If you @mention another agent in a direct chat, the chat becomes a group chat with its recent history. Group chats show in the sidebar with the Squad chat pinned first, and you can rename, archive, and restore them. See Start a group chat.
- Add any MCP server from Settings > Custom integrations > Add MCP server. Paste the server’s URL or the JSON config from its docs, and its tools become tools for your agents. You can also paste the config to an agent in chat. See Add an MCP server or an API key.
- MCP servers that use a browser sign-in now connect with a sign-in window. There is no key to paste. Squad renews the sign-in for you when the server allows it. See Add an MCP server.
- The Memory screen now works. Read and edit About you, each agent’s Notes, Personality, and Identity files, and the squad files (Company, Contacts, Glossary, Squad, and Voice). Search, edit, or delete the memories your squad collected. A meter on About you and on each agent file shows how much of the file the agent reads in each run. See Manage what your squad remembers.
- The new Tools screen in Settings shows every tool your agents can use, including the tools of your MCP servers. You can turn off research tool families and single tools of your MCP servers for the whole workspace. See Turn tools on or off.
- Skill packs are on the Skills screen. Turn ready-made packs on or off, turn off single skills in a pack, or add a pack from a public GitHub repository. Agents can manage packs for you from chat too. See Turn a skill pack on or off.
- Each agent now has its own web browser. Agents open pages, take screenshots, click, and type. They can send you a screenshot in chat.
- Agents can put cards in chat that you act on in place: pick an option, connect an app, add an MCP server, save a key to the vault, or answer a ticket. See Answer chat cards.
- Agents can now start a conversation with you in dashboard chat on their own, for example to report the result of a scheduled job.
- New models: Grok 4.6, the new default for new Grok workspaces, and GLM-5.3 and GLM-5.3 Flash (Ox Alpha) for Z.AI. GLM-5.3 Flash (Ox Alpha) is the new default for new Z.AI workspaces. See Choose models and manage AI accounts.
- Reasoning effort has a new level, Max (GPT-5.6 family), for the GPT-5.6 models. Your reasoning effort setting now also applies to runs on a ChatGPT subscription. Before, those runs ignored it. See Set the workspace model settings.
Improved
- Your own email domain now has its own screen: Settings > Email Domains. It shows the DNS records to add, the status of each record, and a button to check again. See Connect your own domain.
- Telegram is no longer part of setup. You pick your AI provider and launch. You can connect a Telegram bot later in Settings. See Connect Telegram.
- A message that you send while the agent is busy now shows QUEUED. Select Steer now to hand it to the run in progress. See Send a message while the agent is busy.
- Long runs no longer stop because of a time limit. Squad now stops a run only when it shows no progress for 15 minutes, and the run says why it stopped. See Time limits.
- A run’s detail now shows the Reasoning level the run actually used. See See what your agents did.
- You can now name your threads and drag the side panels to make them wider or narrower. The mission board scrolls sideways instead of squeezing its columns.
- Hover over an agent in the sidebar and select the message button to open a direct chat with it.
- The thread list loads older conversations with Show older conversations, so a busy conversation no longer hides the others. Long threads load more with Show earlier replies.
- In the home composer, type
/to pick a skill. - You can now turn off the Deliverable ready email. It is off by default, and the in-app notification stays on. See Choose your notifications.
- When a run fails, for example because your AI connection expired, the chat now shows the real reason and a button to reconnect. A turn that ended without a reply shows a short notice with ↻ Ask again.
- Agents now click buttons and links by their visible name, so their clicks land on the right control. They can also take a screenshot of a whole page, and they close cookie banners with the most private choice before they take screenshots.
- A skill pack now turns on even when one of its skills has the same name as a skill your squad already has. The clashing skills in the pack stay off.
- The Tools screen shows how many MCP tools are on and off.
- When you save Telegram or Discord bot settings, Squad now checks that the change took effect. If it cannot confirm it yet, you see a warning.
- When an MCP server gives a sign-in that expires and cannot renew, Squad now connects it and tells you when it expires. Before, the connection was refused.
- Tables in chat, missions, and docs now scroll sideways instead of breaking words into single letters.
- The dashboard now calls your lead agent the AI Chief of Staff.
- On the Email screen, the Compose button reads Create inbox when no agent has an inbox yet, and it opens inbox creation.
Fixed
- Grok and ChatGPT connections no longer expire on busy agents. The AI Tokens card now shows an expired connection even when only some agents have it.
- MCP servers with more than 64 tools now give your agents all of their tools.
- Long emails no longer arrive empty for your agents.
- Skills that your squad wrote show their name and description on the Skills screen again.
- The Squad chat always shows in the sidebar, also before its first message.
- The first message of a long thread now shows as the thread’s start, even when the thread has many replies.
- Agents’ skills reach every run again. Before, a long list of skills cut some of them off.
- New workspaces finish setup again. A problem in mid-August stopped some of them. It was fixed on August 19.
- Chat headers no longer break agent names into a vertical column when the window is narrow.
- Two-word @mentions, for example of an agent with a first and last name, now show as one mention.