1.8.0

1.8.0: OpenSwarm 1.8: panels beside the canvas, a rebuilt browser agent that signs in without seeing your passwords, chat steps that say what they did, and Sonnet 5.5 for new Claude chats.

New

• Work in panels as well as on the canvas: chats, browsers, apps, files and terminals open side by side as tabs you can drag, and switching between panels and the canvas keeps your pages, tabs and drafts.
• The sidebar has Pinned, Projects and Recents, with pin and archive on hover; on a Mac it is frosted glass over your desktop.
• The browser agent was rebuilt: it fills a form in one step, reads text inside frames, answers a page's pop-ups, and counts a message or a form as sent only when the page shows it, so a message goes out once.
• Browser cards have tabs: hold Ctrl and press Tab to pick one, pin a tab down to its icon, mute one that plays sound, and middle-click to close.
• Agents can sign in with logins you choose to save for them, imported from Chrome or a password manager's export, or saved when you sign in on a card yourself. They fill a login only on its own site and never see the password, and it stays off until you turn on Sign in with logins I saved for agents in Settings > Privacy.
• With your OK, the site sign-ins from a Chrome profile you pick come into the agents' own browser, so sites you are signed in to in Chrome are signed in for agents too.
• Every step in a chat says what it did, a group of steps gets a short title, and rich answers draw the first time far more often, with a comparison card for two to four options.
• New Claude chats start on Sonnet 5.5, and Fable 5.1 works with a Claude subscription, not only an API key.
• A usage limit is named from the provider's own meter, and the chat picks up by itself when it resets.
• Memory keeps a fact only when an agent decides it is worth keeping, and nothing is gathered in the background. New installs start with memory off (Settings > Memory); an install whose memory holds a real fact keeps it on, and stored notes that were never real facts are set aside, where you can look them over or delete them.
• OpenSwarm confirms your account's access at launch and every few hours, and keeps working offline for up to 7 days after the last confirmation; without one it shows Can't confirm your access, with Check again.

Fixed

• The Save your sign-in for agents prompt appears only when Sign in with logins I saved for agents is on, nothing you type is kept while it is off, and Save no longer switches that setting on.
• Browser cards no longer spin forever after a Chrome that closed unexpectedly left its profile locked, and a card waiting on the sign-in window says so and opens by itself once it closes.
• Before changing or deleting a scheduled workflow or one of its steps, the agent checks with you on a card, and a deleted workflow goes to Trash, where it can be restored.
• Stop ends the command the agent was running instead of leaving it running on its own, and a Stop while a tool waits for your approval never runs it.
• Settings sent to OpenSwarm at launch and on subscription changes no longer include your API keys, sign-in tokens, saved instructions or snippets.
• The Claude Code engine inside the app no longer sends its own usage telemetry, error reports or update checks.
• The model router no longer jams and restarts after a long day of heavy use.
• A chat repeating the same step with the same result is stopped at the fifth repeat on one card; a deliberate wait, like a poll, is not.
• Updated built-in Python and secure connections let sites that refused the old ones, Wikipedia among them, load directly.
• Chats hold still: no jump when a reply or browser run ends, and no half-typed formatting while text streams.