Vibe Deck

Why does Windows show a "Windows protected your PC" warning when I install Vibe Deck?

That SmartScreen message appears because Vibe Deck comes from an independent developer and the installer is not code-signed yet - Windows shows it for any new app that has not built up download reputation, not because anything is wrong with the file. To install, click "More info" and then "Run anyway". You can confirm the file is intact by matching its SHA-256 hash to the one shown on your download page. Code signing is on the roadmap and will clear the warning over time.

Is Vibe Deck the best terminal for Claude Code?

Vibe Deck is purpose-built for Claude Code: it runs up to nine live Claude Code sessions in one window, tiles them automatically, and gives you a sidebar that shows what each agent is doing. If you run more than one Claude Code agent at a time on Windows, it is the most direct setup - a plain terminal or tmux can do parts of this but neither is designed around AI coding agents.

What is the best AI terminal?

The best AI terminal is the one built around running AI coding agents rather than generic shells. Vibe Deck is an AI-agent terminal: a single Windows app for orchestrating multiple Claude Code sessions with per-agent status, tiling layouts, and no subscription.

How do I run multiple Claude Code terminals at once?

Install Vibe Deck on Windows, open it, and start up to nine Claude Code sessions inside the one window. Each session is a live terminal; the layout buttons tile them into Grid, Columns, or Focus, and the sidebar tracks what each agent is working on. No tmux config or separate windows required.

Does Vibe Deck require a subscription?

No. Vibe Deck is a one-time purchase - pay once and own it, with free updates across the current major version.

What operating systems does Vibe Deck support?

Vibe Deck runs on Windows 10 and Windows 11 today. Linux is coming soon - join the waitlist on this page.

• Stability overhaul: 30+ verified fixes, per-terminal native Chrome, and deep agents
• Native Claude-in-Chrome per terminal: right-click a Claude terminal and toggle Native Chrome (--chrome) to give that one agent the real signed-in Chrome; everyone else keeps the built-in web pane. One Chrome connection per machine, so it is a per-terminal choice, applied on restart.
• Deep agents: say "launch two deep agents" in any terminal and the agent becomes an orchestrator, delegating to autonomous deep-agent workers that plan, use the filesystem as shared memory, run in parallel, and return synthesized results. Ships as a bundled skill + subagent installed into every profile.
• Cards and pane bars now show your last submitted prompt instead of whatever title the agent happened to set, so a wall of terminals reads like a to-do list. Non-Claude terminals keep their normal titles.
• Crash fix: closing a browser tab (or any browser cleanup) while a page was still loading could take down the entire app. It no longer can.
• Safety fix: turning Remote Control on used to press Enter in every Claude terminal, which could silently confirm a pending permission dialog. Injection now waits until each terminal is idle.
• Your deck can no longer factory-reset itself: a locked or torn config file used to be silently replaced with defaults; config, MCP config, and agent rules are now written atomically and recover from transient failures, and backup pruning keeps the newest backups instead of sometimes deleting them.
• Conversations survive: fresh accounts no longer lose transcripts to the per-terminal cleanup, resume ids are no longer dropped for terminals still waiting in the spawn queue, and folder trust answers, MCP server approvals, and prompt history now stick across relaunches. The Chrome onboarding answer finally sticks on the default profile too.
• Agent routing survives restarts: toggling Chrome, switching accounts, or disabling Remote Control used to silently strip a terminal's model-provider setup (custom command, args, base URL, key). Duplicated workspaces now carry accounts, providers, notes, and the Chrome flag as well.
• Vibe Loops: loop agents' browsers were being cleaned up as "stale" within 60 seconds, and closing the workspace a loop happened to start in killed the loop. Both fixed.
• Usage meters: the Claude usage poll can no longer hang forever on a bad connection, and a provider reporting 1% used no longer renders as a full red 100% bar.
• Browser tools hardened: waits can no longer hang forever mid-navigation, an agent at the browser limit gets a clear error instead of silently destroying another agent's browser, and cookie clearing now requires a site (no more one-call wipe of every login).

• Snip and Annotate, agents that cannot quit on you, and a leaner agent browser
• Snip and Annotate: a new capture-and-markup tool. Trigger it from the Snip button on the left toolbar, a configurable global hotkey, Ctrl+Shift+S, or the Ctrl+K menu - the screen freezes so you select an Area or the Full screen against a still frame. The shot opens in a separate, movable editor window (the deck stays live) with arrows, boxes, ellipses, lines, freehand, text, a numbered-step counter, a highlighter, undo/redo, one-click "beautify" gradient backgrounds, and a redaction tool that destroys the pixels on export so secrets never leave your machine. Insert the result straight into a terminal, copy it, or save it; recent shots stay in a history strip.
• Agents can no longer shut down Vibe Deck: closing the window, Alt+F4, Cmd+Q, or a menu quit now asks for confirmation while any agent terminal is live, one agent can no longer type a kill command into another terminal, and every agent is told it must never quit, close, or kill the app or its terminals. Ending a session stays your decision.
• Fewer captchas for your agents: each agent's private browser now presents as plain Chrome instead of advertising Electron, which cuts the extra captchas and sign-in walls that bot-detection used to throw.
• Cheaper, more reliable agent web work: agents now read a page as clean markdown plus its structured data - a fraction of the tokens of a full snapshot or screenshot - screenshots come back as small downscaled images and no longer hang on heavy pages, and page snapshots can be scoped so a big page does not flood an agent's context.
• Fixes: agent browsers no longer steal focus when they open, a browser left open by an agent can be closed from the pane, and the "press Ctrl+A" prompt detection no longer misfires.

• Minimize to a tray, drag to resize, and agents that talk to each other
• Minimize terminals to a tray: park a terminal you are not using in a dock at the bottom of the deck. It keeps running and keeps its full scrollback, and the remaining terminals re-tile to fill the freed space. Bring it back from its tray chip, its sidebar card, Ctrl+number, or Ctrl+Shift+M; "Minimize others" solos the terminal you are in.
• Drag to resize terminals: drag the divider between any two panes to give one more room and another less. Double-click a divider to even them back out. Your split is remembered per workspace and per layout.
• Agent messaging (opt-in): turn on "Agent messaging" in the Agent Tools panel and the Claude terminals in your deck can coordinate. Each can post a one-line note of what it is working on, see what the others are doing, and send messages to another terminal's inbox. Messages are delivered when the receiving agent is free, so they never interrupt it mid-task, and the whole inbox is cleared when you close the app.
• Terminal ids on every pane: each terminal now shows its id (s1, s2, ...) on its header and sidebar card - the same handle you use when you tell one agent to message another ("send s2 the API schema").
• Screen recording for agents: agents can now record the screen or a single window to a video you can replay, and get sampled still frames back so they can see what happened over time.
• Restore your default Claude account: a one-click way to bring back the default account if you had hidden it from the Accounts panel.
• Fix: the layout picker menu no longer runs its text together - each layout's name and description now sit on their own lines.

• Prompt library, Remote Control, and a calmer, more navigable terminal
• Prompt library: save and reuse your best prompts. Open it from the new library button in the sidebar header or with Ctrl+Shift+P, search by title, body, or tag, and copy any prompt to the clipboard in one click. Prompts persist locally and ride along in your setup export/import.
• Remote Control: one sidebar toggle launches every Claude terminal with Claude Code's --remote-control flag so the session is drivable from claude.ai and the Claude app. Flipping it re-initiates your running Claude terminals carrying their conversation, so nothing loses context.
• Ctrl+click a file path in any terminal to reveal it in the file manager - the same gesture as Ctrl+click on a URL. A file is revealed selected (never executed), a folder opens; relative paths resolve against the terminal's folder.
• Prompt rail + jump: a slim tick gutter down the right edge of each terminal marks every prompt you have sent. Click a tick to jump straight to that prompt (hover to see its text), or step between prompts with Ctrl+Up / Ctrl+Down.
• Spotlight: in a multi-pane workspace the quiet terminals gently dim so your eye lands on the one that is working or waiting on you.
• Easier scrolling: a "jump to latest" pill appears with an unread count whenever you scroll up, the view no longer strands output after a resize or layout change, the scrollbar is wider and easier to grab, history is deeper (5000 lines), and Alt+scroll flies a page per notch.
• Setup backup moved into the gear menu: Export setup and Import setup are now right where you would look for them, instead of buried in the voice panel.
• A calmer terminal palette and more legible dimmed text make dense agent output easier to read.

• Private browsers for every agent, real CDP automation
• Each agent now gets its own private browser that other agents cannot touch. Call browser_open to get a handle like "vb-3" and pass it to every browser_* tool. Sub-agents also open their own - no more terminals fighting over the same page.
• 25 browser tools powered by real CDP: trusted keyboard, mouse, and drag events replace synthetic DOM injection, so clicks and typing work on every site.
• browser_snapshot returns a labeled outline with each element's live value (typed text, checked state, selected option), pierces shadow DOM and same-origin iframes, and surfaces non-semantic clickables with include_all.
• Element-crop screenshots: pass a uid from a snapshot to browser_screenshot to capture just that element. browser_resize sets any viewport for responsive testing.
• Agents browse headlessly in the background - the Web pane no longer pops open automatically. The Web button now shows a count badge (pulses while loading) and a hover popover listing every agent's open tab; click a row to watch it.

• A calmer, smarter deck: workspace tabs, clean boots, live pane bars
• Workspace tabs in the titlebar: browser-style tabs with color dots, activity badges, and session counts. Click to switch, drag a session card onto a tab to move it. The sidebar now belongs entirely to your session cards.
• Terminals boot clean: no more launch-command flag soup flashing before the agent banner - every session opens straight on the agent.
• Pane bars tell the live story: each terminal's bar shows what its agent is working on, which Claude account it runs on, and its token burn meter. Past 50% context the meter spells it out ("412k tok - 56% ctx").
• Drag a folder onto the sidebar and a terminal opens right there, named after the folder - drop several and each gets its own.
• One consistent toolbar (Grid, Voice, Web, Tools, Rules, settings gear) replaces the old stack of mixed buttons; the gear gathers voice settings, accounts, usage, shortcuts, and the config file.
• Multi-account usage is now one compact row per account with session/week mini bars and a green arrow on the account with the most headroom; the switch button only appears when there is actually somewhere better to switch to.
• Quieter visuals: pane selection is a gold edge on the pane bar instead of a full border, and fresh terminals show a faint hint that teaches drag & drop and snapshots, gone on your first keystroke.

• Show your agents anything: snapshots, drag & drop, live token burn
• Snapshot tool: a scissors button on every terminal lets you capture any region of your screen - the file path pastes instantly so the agent can look at exactly what you saw, with a preview card confirming what was attached.
• Drag & drop images, files, or folders onto any terminal: paths paste ready to send, every item gets a card (thumbnails for images, type chips for files), and clicking a card copies it for another terminal. Pasting images straight from the clipboard works too.
• Smart attachment cards: deleting a pasted path or [Image #N] token in the prompt removes exactly that card - they track what the terminal actually shows.
• Live token burn chip on every Claude pane: total session tokens with a context-fill bar that goes gold, amber, then red as the window fills. Hover for in/out tokens, context percentage, and session cost. Resets the moment you /clear.
• Session handoff notes: a pencil on each session card opens a sticky note ("what is this agent doing, what should I check?") that survives restarts and restores - no more guessing what terminal 3 was working on yesterday.
• GitHub agent tools: agents can read repo info, check the current branch's CI status, and list, inspect, create, and merge pull requests.
• Agent rules: define global always-do / never-do rules once and every Claude terminal receives them, with a banner that suggests new rules as you work.
• Voice pronunciation dictionary: teach dictation your vocabulary - say "use effect", get useEffect.
• Editor-style Ctrl+A: selects everything you typed or dictated in the inline input; Ctrl+C copies it cleanly, Backspace deletes it.
• Terminal quality of life: find in terminal (Ctrl+Shift+F), Ctrl+click opens links, font zoom (Ctrl +/-), a keyboard cheatsheet (Ctrl+/), inline image rendering, and a copy-whole-terminal button.
• Faster and quieter: usage bars and burn chips refresh every second, and reopening the app no longer triggers voice or notification alerts while restored conversations boot.

• Multi-account that actually works: one-click switching and auto-switch
• Connecting a second Claude account now just works: adding an account opens a sign-in terminal automatically - follow the login prompts once and you are connected. Accounts that are not signed in yet get a "sign in" button.
• One-click account switching: a switch button lives right under the usage bars and on every account row. Switching asks whether to bring each terminal's conversation over to the other account or start fresh - carry-over resumes every terminal exactly where it left off.
• Auto-switch near the usage limit (on by default, can be turned off): when the active account nears either its 5-hour or weekly limit and another account has real headroom, every terminal moves over automatically with its conversation, and you get a notification.
• Honest headroom: accounts are judged by their worst window - an account with a maxed weekly limit is never called "best" just because its 5-hour window is empty. Idle accounts' usage bars stay live too (polled from Claude's own usage endpoint).
• Clearer account UI: the active account is marked with an ACTIVE chip everywhere, accounts can be renamed, and the deck reopens on whatever account you closed it with.
• Resume on restore: closing and reopening Vibe Deck brings every Claude terminal back into its own conversation automatically (toggle in Claude Options, on by default).
• Smarter voice announcements: "done" now waits for sustained quiet (no more false alarms mid-command), and announcements carry context - which terminal finished what, or the actual question on screen.
• Fixed: switching grid/columns/focus layouts could garble terminal visuals; terminals also gain right-click "Redraw terminal" (Ctrl+L) and "Clear scrollback".
• Fixed: agent-tools toggles flipped mid-session were ignored by running Claude sessions.
• Fixed: the SyntroAI logo now shows everywhere on Windows (taskbar, title bar, Explorer menu) instead of the Electron logo.
• Explorer "Open in Vibe Deck" is now registered automatically at every start and handles paths with spaces reliably.

• Multiple Claude accounts and Explorer integration
• Multiple Claude accounts (opt-in): add extra accounts in isolated profiles, pin any terminal to an account from its card's right-click menu, and set a deck-wide default. Each account gets its own 5-hour and weekly usage bars, with the account that has the most headroom highlighted.
• Switch a live terminal to another account without losing your place: Vibe Deck offers to carry the conversation over (resume), start fresh, or cancel.
• Windows Explorer integration: turn on the toggle in Agent Tools and right-click any file or folder to "Open in Vibe Deck" - it routes into the running app (or starts it) and asks whether to use the current workspace or a new one. Per-user setting, no admin needed.
• Version badge: the app version now shows next to the Vibe Deck title, so you always know what you are running.
• Removed: cloned-voice announcements (Chatterbox). Loading the voice model could freeze the machine, so spoken alerts now always use the fast system voice. Voice announcements and the agents' speak() tool are unchanged otherwise.

• Agents with eyes: browser, screen vision, and alerts
• Claude can now open a built-in browser pane, screenshot it, read console errors, and click around - it sees its own work
• Agents can screenshot any window on your machine and (optionally) read your other terminals
• Vibe Deck taps you when an agent needs you: pulsing card, desktop notification, sound, and optional spoken announcements
• Agent tools over a built-in local MCP server, auto-connected to every Claude session you launch and managed from the new Agent Tools panel
• Self-aware sessions: each Claude terminal is briefed on its Vibe Deck environment at startup (which terminal it is, what else is running, what is enabled) and can re-check live with vibe_deck_context
• Screen vision: agents list open windows and screenshot any window or the full screen
• Terminal awareness (off by default): agents list and read your other terminals, with a separate toggle to let them type into other sessions
• Clipboard tools (off by default): agents read and write the system clipboard
• Real Chrome control (off by default): one-click debug Chrome launch wired to chrome-devtools-mcp so agents can drive your actual browser
• Attention alerts: pulsing session card, desktop notification, and sound when an agent needs input, finishes long work, or dies while you are away
• Voice announcements (off by default): the app speaks when an agent is blocked on you, plus a speak() tool agents can use for short updates

• Workspaces: group terminals by feature or project, switch with a click or Ctrl+Tab
• Background workspaces keep their agents running, with live activity badges
• Per-workspace color, layout, and agent command override
• Move sessions between workspaces by drag or right-click Move to
• Coding agent picker: Claude, Gemini, Codex, Grok, or any custom CLI command

• Initial public release
• Grid, Columns, and Focus layouts with drag-to-reorder session cards
• Quick Start launcher with multi-folder project open and Restore Last Deck
• Claude usage meters: live session and weekly limit bars with reset countdowns
• SyntroAI Voice: dictate into the focused terminal via Windows voice typing
• Claude options card: skip-permissions, continue conversation, custom flags