Skip to main content

​
UI Overview

Maestro features a three-panel layout:
  • Left Panel - Agent list with grouping, filtering, search, bookmarks, and drag-and-drop organization
  • Main Panel - Center workspace with two modes per agent:
    • AI Terminal - Converse with your AI provider (Claude Code, Codex, or OpenCode). Supports multiple tabs (each tab is a session), @ file mentions, image attachments, slash commands, and draft auto-save.
    • Command Terminal - PTY shell with tab completion for files, branches, tags, and command history.
    • Views: Session Explorer, File Preview, Git Diffs, Git Logs
  • Right Panel - Three tabs: File Explorer, History Viewer, and Auto Run
Main screen

​
Agent Status Indicators

Each agent shows a color-coded status indicator:
  • 🟢 Green - Ready and waiting
  • 🟡 Yellow - Agent is thinking or waiting for user input
  • 🔴 Red - No connection with agent
  • 🟠 Pulsing Orange - Attempting to establish connection
  • 🔴 Red Badge - Unread messages (small red dot overlapping top-right of status indicator, iPhone-style)

​
Git Actions

For agents whose working directory is a git repository, the same set of git actions is reachable three ways:
  • Header branch pill - hover the pill showing the current branch name (clicking works too).
  • Left Bar right-click - right-click the agent in the agent list.
  • Command palette (Cmd+K / Ctrl+K) - search for the action by name, no mouse required.
The header menu also shows the current branch and origin at the top. Each has a copy button, and clicking the origin opens the repository in your browser. Below the actions it offers Configure Worktrees. The header pill and the command palette act on the agent you’re looking at. The right-click menu acts on the agent you right-clicked, so you can pull or check the log of a background agent without switching to it first. Pull and push stream their output as it happens, so you can watch the transfer and read git’s error message if it fails. Dismissing the modal leaves the command running; Cancel stops it. When a push fails because the branch has no upstream, the modal offers a one-click Push and Set Upstream retry. Working-tree changes are shown by the git status widget beside the pill (+ additions, − deletions, ~ modified). Hover it for a list of changed files with diff bars, plus shortcuts to the full diff and the log.

​
File Explorer and Preview

The File Explorer (Right Panel → Files tab) lets you browse project files. Click any file to open it in the File Preview view. File viewer File Preview features:
  • Syntax highlighting for code files
  • Markdown rendering with toggle between raw/preview (Cmd+E / Ctrl+E)
  • Image viewing for common image formats
  • Audio and video playback with a speed control that sticks (see below)
  • CSV and TSV tables with sortable columns and a per-row detail view (see below)
  • Line numbers for easy reference
  • Search within file (Cmd+F / Ctrl+F)

​
CSV and TSV Tables

.csv and .tsv files render as a real table instead of raw text. Click any column header to sort by it (click again to reverse, a third time to clear), and Cmd+F / Ctrl+F filters the table down to matching rows with the hits highlighted. Wide rows get cut off at the edge of the screen, which is the wrong shape for reading one record closely. Double-click a row to flip it into a vertical view: a modal listing every column as a field/value pair, one per line. Long and multi-line values wrap and keep their line breaks instead of being truncated, and each value has a copy button on hover. Inside that view:
  • Left and Right arrows, or the chevrons in the header, step through rows without closing the modal. Navigation follows whatever the table is currently showing, so it respects your active sort and search.
  • Up and Down arrows scroll the field list, with PageUp / PageDown for whole screens and Home / End to jump to the top or bottom. Handy when a row has more fields than fit at once.
  • / jumps to the filter box, which narrows to fields whose name or value matches what you type. Enter hands focus back to the field list so the arrows resume stepping through rows.
  • Drag any edge or corner to resize. The size is remembered for next time.
  • Esc closes the row view and leaves the file open.
The field list takes focus the moment the view opens, so every one of those keys works without clicking first.

​
Audio and Video Playback

Click an audio or video file and it opens in a player right inside the File Preview. Supported formats are the ones Chromium can decode: mp3, wav, m4a, aac, flac, ogg, oga, opus, weba for audio, and mp4, m4v, webm, mov, ogv for video. Anything else (mkv, avi) still gets the “Open in Default App” fallback. Playback starts as soon as you open the file, and keeps going when you switch tabs or agents. Start a podcast, go work somewhere else, and it follows you.

​
The Floating Player

When you browse away from the tab, the player detaches into a small floating widget that stays on top of whatever you are doing:
  • Drag the title bar to move it anywhere, and drag the bottom-right corner to resize it. It remembers where you left it, across restarts. Double-click the corner to reset the size.
  • Minimize it to a slim pill with just the title and a play/pause button.
  • Double-click the title bar (or click the filename) to jump back to the file’s tab, which re-docks the player into it.
  • Close it to get it out of the way. This hides the controls only - the audio keeps playing. Bring it back by opening a media file again, or with Show Floating Media Player in the command palette (Cmd+K / Ctrl+K).
Audio-only files get just the controls, since there is no picture to show.

​
One Player at a Time

Only one file plays at a time, so you never end up with two podcasts talking over each other. You can still have several media files open: the prev/next buttons in the transport step between them, and each one remembers its position, so you can jump back and forth without losing your place. Files stream from disk with range requests, so scrubbing a multi-gigabyte screen recording is instant and does not load the file into memory. Playback speed persists. Whatever rate you pick (anywhere from 0.25x to 4x) carries over to the next file you open and survives a restart. Pitch is preserved, so a 2x podcast still sounds like a person. You can also set the speed outside the app with maestro-cli settings set mediaPlaybackRate 1.5.

​
File Explorer Keyboard Shortcuts

With the Files tab focused, navigate the file list without touching the mouse: Use Shift+Up / Shift+Down when you want to glance further down the tree without losing your place - the focused file stays put while the viewport slides. When you open a file, a breadcrumb trail appears showing your navigation history. Click any breadcrumb to jump back to a previously viewed file. This makes it easy to compare files or return to where you were.

​
File Editing

Files can be edited directly in the preview. Press Cmd+S / Ctrl+S to save changes. If you navigate away or close the preview with unsaved changes, a confirmation dialog will ask whether to discard them.

​
Drag and Drop Files Into the Tree

Drag files or folders from your OS file manager (Finder on macOS, Explorer on Windows, your file manager on Linux) straight into the Files tab to copy them into the project:
  • Drop onto a folder row to copy the items inside that folder. The folder highlights as you hover it.
  • Drop on empty space or a file row to copy into the project root. The panel border highlights to show the root is the target.
  • Folders are copied recursively, with all of their contents.
  • If a name already exists at the destination, a prompt lets you Overwrite, Auto-rename (adds a numeric suffix), or Skip the conflicting items.
Where you drop decides what happens:
Importing into the tree copies from your local machine, so it is not available for agents running on an SSH remote. Attaching files to the chat still works on remotes.

​
Publish as GitHub Gist

Share files directly as GitHub Gists from the File Preview: Prerequisites:
  • GitHub CLI (gh) must be installed
  • You must be authenticated (gh auth login)
To publish a file:
  1. Open a file in File Preview
  2. Click the Share icon (↗) in the header toolbar, or
  3. Use Cmd+K / Ctrl+K → “Publish Document as GitHub Gist”
Visibility options: The confirmation modal focuses “Publish Secret” by default, so you can press Enter to quickly publish. Press Esc to cancel. After publishing:
  • The gist URL is automatically copied to your clipboard
  • A toast notification appears with a link to open the gist in your browser
The share button only appears when viewing files (not in edit mode) and when GitHub CLI is available and authenticated.

​
@ Mentions

Reference files in your AI prompts using @ mentions:
  1. Type @ followed by a filename
  2. Select from the autocomplete dropdown
  3. The file path is inserted, giving the AI context about that file
The same @ picker can also reference other agents. Alongside files, it has an Agents section - pick one to consult it inline and stream its reply back into your chat. Maestro tells files and agents apart by shape: a path-like @src/app.ts is a file, while a bare @codex is an agent.

​
Command Mode (!)

Press ! in an empty chat composer and it turns into a command line: what you type next runs as a shell command in the agent’s working directory instead of being sent to the agent. It is a way to check something without leaving the chat. The ! is a gesture, not part of the command - it disappears the moment it switches modes, and you just type the command:
You will know you are in command mode: a $ appears at the left of the input, and a COMMAND MODE strip above it names the directory the command will run in. Getting back to the agent: press Esc on an empty command line (or Backspace, same thing). The composer keeps focus, so you can carry straight on typing your message. Command mode sticks around between commands, so you can run several in a row without retyping !, and you leave deliberately when you are done. How it behaves:
  • The agent is bypassed entirely. It is never spawned, never written to, and never sees the command or its output. Nothing you run this way enters the agent’s context - if you want the agent to see the result, copy it into a message.
  • It runs immediately, even while the agent is working. Command mode does not queue and does not interrupt the turn in progress, so you can check git log while the agent is mid-edit.
  • It runs in the agent’s working directory (on the agent’s SSH remote, if it has one). Each command is independent - there is no persistent shell, so cd src on its own does nothing. Chain instead: cd src && ls.
  • Every command gets its own card, never merged into the surrounding conversation. The card shows the command, where it ran, and a live spinner while it works; when it finishes, the exit code and how long it took.
  • Colour is preserved. Output keeps the colours the command produced (git status, eza, rg), rendered properly rather than shown as raw escape codes. The copy button gives you clean, uncoloured text.
  • The draft survives a tab switch, mode and all. Leave a half-typed command, go read another tab, come back, and it is still a command.

​
Tab Completion in Command Mode

Command mode gets the same Tab completion the Command Terminal has, so you are not typing paths from memory: One match completes in place. Several open a picker: ↑ / ↓ to move, Enter to accept, Esc to dismiss. In a git repo, Tab inside the picker cycles the category filter (All, History, Branches, Tags, Files) and Shift+Tab cycles back. Completion resolves from the agent’s working directory, which is where the command will actually run. This is deliberately not the Command Terminal’s directory - cd-ing in a terminal tab does not move where your commands run, so it must not move where completion looks either. Chat affordances stand down in command mode, because a shell line means different things by the same characters:
  • @ file mentions - an @ here is an scp target or an email in a commit message, not a file reference for the agent.
  • Slash commands - a leading / here is an absolute path (/usr/bin/env), not /history.
  • Image attachments - pasting or dropping an image is ignored, and the attach button is hidden. A shell command has nothing to do with an image, and the agent (which is the thing that reads them) never sees this input.
Stopping a command: a running command’s card has a Stop button. Reach for it if you start something long or something that waits for input.
Command mode has no keyboard - nothing is connected to the command’s stdin. Programs that prompt for input (sudo, an editor opened by git commit, an interactive installer) will hang until you press Stop. Run those in a Command Terminal tab instead.
Very large output is capped so a runaway command cannot bloat your transcript; the card says so when it truncates. Sending a message that starts with !: typing ! first only enters command mode when the composer is empty, so a bang inside a sentence is safe. To start a message with one, prefix it with a backslash: \!important reaches the agent as !important. Command mode is AI-chat only. In a terminal tab or the legacy terminal mode you are already at a shell, so ! is an ordinary character.

​
Prompt Composer

For complex prompts that need more editing space, use the Prompt Composer - a fullscreen editing modal. To open the Prompt Composer:
  • Press Cmd+Shift+P / Ctrl+Shift+P, or
  • Click the pencil icon (✏️) in the bottom-left corner of the AI input box
Prompt Composer Button The Prompt Composer provides:
  • Full-screen editing space for complex, multi-paragraph prompts
  • Character and token count displayed in the footer
  • All input controls - History toggle, Read-only mode, Thinking toggle, and send shortcut indicator
  • Image attachment support via the image icon in the footer
Prompt Composer When you’re done editing, click Send or press the displayed shortcut to send your message. The composer closes automatically and your prompt is sent to the AI.

​
Message Queue

You never have to wait for an agent to finish before lining up your next thought. Any message (or slash command) you send while an agent is busy is added to that tab’s queue instead of being dropped, and dispatched automatically, in order, the moment the agent becomes ready. A QUEUED (n) separator appears in the transcript listing everything waiting to go. Each queued item has a row of controls (hover reveals them, and they stay visible while you work): Editing is only offered for messages, not slash commands. Long messages are truncated with a Show all toggle so the queue stays compact, and attached images collapse behind a click-to-expand thumbnail strip (click a thumbnail to open it full-size in the carousel).

​
Force Send

When Forced Parallel Execution is enabled and another tab in the same agent is already working, a queued item gains a Force Send button. This dispatches that message immediately, running it in parallel instead of waiting its turn in the cross-tab queue. A confirmation lists which other tabs are currently busy before it sends.

​
Execution Queue view

Press Cmd+Shift+X / Ctrl+Shift+X (or click the queue indicator) to open the Execution Queue - a single view of everything queued across all of your agents. It offers the same per-item controls (edit, copy, hold/resume, reorder, remove) plus a jump-to-agent shortcut, so you can manage a busy fleet from one place. Items are processed sequentially per agent to keep concurrent file edits from colliding.

​
Input Toggles

The AI input box includes three toggle buttons that control session behavior: Input Toggles Per-tab persistence: Each toggle state is saved per tab. If you enable Thinking on one tab, it stays enabled for that tab even when you switch away and back. Permission mode and ask-back questions: Claude Code agents also offer a permission mode you cycle by clicking the permission pill in the input toolbar (Full Access -> Standard -> Read-Only). In Standard mode, tool calls that need approval and any AskUserQuestion ask-backs appear as in-app prompts you answer inline. In Full Access mode there are no prompts at all: permission checks are bypassed, and an AskUserQuestion call cannot be answered, so the turn stalls until you stop the agent. Sending a follow-up message won’t unstick it: the message only queues behind the stalled turn, which never completes to dispatch it. Avoid instructing agents to ask questions through tools unless the tab is in Standard mode. See provider notes for the full explanation.

​
Configuring Defaults

Set the default state for new tabs in Settings (Cmd+, / Ctrl+,) → General: Input Toggle Defaults

​
Send Key Configuration

Configure how messages are sent in each mode:
  • When set to Cmd+Enter / Ctrl+Enter, pressing Enter alone creates a new line (for multi-line input)
  • When set to Enter, use Shift+Enter for new lines
  • The current send key is displayed in the input box (e.g., ”⌘ + Enter”)
  • Per-tab override: Click the send key indicator in the input box to toggle between modes for that tab
When working with image attachments, use the Image Carousel to view, manage, and remove images. To open the Image Carousel:
  • Press Cmd+Y / Ctrl+Y, or
  • Click the image icon in the input box when images are attached
Carousel controls:
  • Arrow keys - Navigate between images
  • Delete or Backspace - Remove the currently selected image
  • Click the X - Remove an image by clicking its remove button
  • Esc - Close the carousel
Images can be attached via drag-and-drop, paste, or the attachment button. The carousel shows all images queued for the current message.

​
Output Filtering

Filter and search through AI output to find specific content or hide noise.

​
Global Filter

The global filter applies to all AI output in the current session. To open the global filter:
  • Click the filter icon in the output toolbar
  • The filter bar appears at the top of the output area

​
Per-Response Filters

Each AI response has its own local filter. Hover over a response to reveal the filter icon, then click to open the filter bar for that specific response.

​
Filter Modes

Click the mode icon to toggle between Include and Exclude.

​
Text vs Regex Matching

Click the indicator to toggle between plain text and regex mode.

​
Filter Controls

  • Query input - Type your search term or regex pattern
  • Esc - Clear the filter and close the filter bar
  • Click outside - If the query is empty, the filter bar closes

​
Placeholders

The placeholder text updates to reflect the current mode:
  • “Include by keyword” / “Exclude by keyword” for plain text
  • “Include by RegEx” / “Exclude by RegEx” for regex mode

​
Use Cases

Finding specific content:
  • Set to Include mode with plain text
  • Type a keyword like “error” or “function”
  • Only matching lines are shown
Hiding verbose output:
  • Set to Exclude mode with plain text
  • Type patterns like “debug” or “verbose”
  • Matching lines are hidden from view
Complex pattern matching:
  • Enable Regex mode
  • Use patterns like \berror\b for word boundaries
  • Or ^\s*# to match comment lines

​
Searching Message History

Two searches share the magnifying-glass menu in the tab bar, and both accept plain text or a regular expression (toggle the Aa / .* chip).

​
In the current tab

Cmd+F / Ctrl+F opens the Find bar over the conversation you’re looking at. Every match is highlighted inline; Enter and Shift+Enter step forward and backward through them.

​
Across every open tab

Opt+Cmd+F / Alt+Ctrl+F opens a modal that searches the message history of all open tabs in the current agent at once. It’s also in the command palette as “Search: Messages (All Agent Tabs)”. Results are grouped by tab, each row showing who said it, when, a preview of the hit with the match highlighted, and a pill when that message contains several matches. The tab you’re currently on is labeled “current”. Pick a result with Enter or a click and Maestro:
  1. Switches to that tab
  2. Scrolls to the message and flashes it so you can see where you landed
  3. Seeds that tab’s Find bar with the same query, positioned on the match you picked, so Enter / Shift+Enter continues from there
Navigate the list with the arrow keys, PageUp / PageDown, and Home / End. Very broad queries are capped so a single character can’t stall the UI: 100 matches per tab and 500 overall. When a search hits either limit the modal says so, and narrowing the query brings the rest into view.
This search covers the AI tabs of the agent you’re currently on, not your whole fleet. Group chats have no AI tabs to search, so the menu entry is hidden there.

​
Command Interpreter

The command interpreter can be focused for a clean, terminal-only experience when you collapse the left panel. Command interpreter

​
Command Terminal

Each agent has a Command Terminal alongside its AI Terminal - a real PTY shell scoped to the agent’s working directory. Switch between them with Cmd+J / Ctrl+J. Open a new terminal tab with Ctrl+Shift+ + `; close, rename, and reorder it just like an AI tab. Right-click (or hover) a terminal tab to open its action menu.

​
Startup Command

Configure a command to run automatically every time a terminal tab’s shell is started - including after you quit and reopen Maestro. This is the simplest way to keep something like npm run dev, a watcher, or a long-running log tail attached to a specific tab. To configure:
  1. Hover the terminal tab and open its action menu.
  2. Click Startup Command… (right under Rename).
  3. Enter the command and, optionally, a working directory. The working directory defaults to the agent’s working directory if left blank.
  4. Click Save.
Behavior:
  • The command runs each time the PTY for that tab is spawned. The most common trigger is launching Maestro after a quit - any open terminal tab is restored, its shell respawned, and the configured command executes.
  • Configuring a command on an already-running shell does not retroactively run it. The next spawn (app restart, or close-and-reopen the tab) picks it up.
  • The configured working directory becomes the shell’s spawn directory, so the command starts in the right place even if the tab’s last cd was somewhere else.
  • Leave the command field empty and save to disable the feature for that tab.
  • Each terminal tab has its own startup command - one tab can run a dev server while another runs a log tail.
SSH agents: when the agent is configured to run on a remote host, the terminal tab also runs on that host, and the startup command executes remotely (the working directory must be a path on the remote machine).

​
Agent Management

Agents are the core of Maestro - each agent represents an AI coding assistant running in its own workspace.

​
Creating Agents

To create a new agent:
  1. Press Cmd+N / Ctrl+N, or click the New Agent button in the bottom-left sidebar
  2. Choose Manual Setup or Guided Setup (Wizard) - see Getting Started for details on each path
  3. For Manual Setup: select an available AI provider (Claude Code, Codex, OpenCode, or Factory Droid), choose a working directory, and optionally name the agent
Advanced configuration options:
  • New Session Message - A hidden message prefixed to the first message whenever a new session (tab) is created. Use this for initial context, setup instructions, or persona definitions that should apply at the start of every conversation. Not visible in chat.
  • Nudge Message - A hidden message appended to every interactive user message sent to the agent. This is useful for persistent instructions or reminders that guide the agent’s behavior across all conversations. Note: Nudge messages only apply to interactive AI messages - they are not included in Auto Run tasks.
  • Custom Path - Override the default executable path
  • Custom Arguments - Additional command-line arguments
  • Environment Variables - Custom environment variables for the agent process
  • Model Selection - Choose a specific model and (where supported) reasoning/effort level. This sets the default for new tabs in this agent. You can override the model or effort on any individual tab using the model/effort pill in the input bar - per-tab overrides only affect that tab and don’t change the agent default or any other tab.
  • Additional Directories - Grant the agent access to directories beyond its working directory. Add a row per directory, then toggle R (read) and W (write) independently: a directory can be read-only reference material, a write-only drop box the agent should never read back, or both. A row with neither toggle lit is inert and is not sent to the agent. Each row also takes an optional description - a short hint about what the directory is for or how the agent should use it, which is passed to the agent alongside the access rule. Providers that support directory flags (for example Claude Code’s --add-dir) also receive these grants natively; the read/write split and the descriptions are always carried in the agent’s system prompt.

​
Editing Agents

Right-click any agent in the left panel and select Edit Agent… to modify its configuration. You can change the name, new session message, nudge message, custom paths, arguments, environment variables, additional directories, model, and effort. Model and effort set here apply as the default to new tabs; existing tabs that haven’t been overridden also follow this default. To override on a single tab without changing the agent-wide default, use the model/effort pill in that tab’s input bar.

​
Deleting Agents

Right-click an agent and select Remove Agent to delete it. This removes the agent from Maestro but does not delete any files or AI session data.

​
Agent Configuration via Quick Actions

Use Cmd+K / Ctrl+K → “Edit Agent” to quickly access agent configuration for the current session.

​
Left Panel Operations

The left panel (sidebar) contains your agent list, groups, and navigation controls. Press Cmd+F / Ctrl+F while the sidebar is focused to open the session filter. The filter:
  • Searches agent names and AI tab names
  • Automatically expands groups containing matches
  • Shows matching bookmarked agents
  • Searches worktree branch names

​
Bookmarks

Pin important agents to the top of the list:
  • Right-click an agent → Add Bookmark
  • Or use the context menu to toggle bookmark status
Bookmarked agents appear in a collapsible “Bookmarks” section at the top of the left panel.

​
Groups

Organize agents into groups for better project management: Creating groups:
  • Cmd+K / Ctrl+K → “Create Group”
  • Groups have a name and emoji for visual identification
Moving agents to groups:
  • Right-click an agent → Move to Group → Select target group
  • Or drag-and-drop agents between groups
Collapsing/Expanding:
  • Click the group header to collapse or expand
  • Groups remember their collapsed state

​
Drag and Drop

Rearrange agents by dragging them:
  • Drag agents between groups
  • Drag to reorder within a group
  • Drag to the “Ungrouped” section to remove from a group

​
Context Menu

Right-click any agent for quick actions:
  • Rename - Change the agent’s display name
  • Edit Agent… - Open configuration modal
  • Add/Remove Bookmark - Toggle bookmark status
  • Move to Group - Organize into groups
  • Move to Window - Send the agent to another Maestro window
  • View Git Log / View Git Diff / Git Pull / Git Push / Change Branch / Create Pull Request - the full git menu, for git repositories only
  • Create Worktree - Create a git worktree sub-agent (if configured)
  • Configure Worktrees - Set up worktree configuration
  • Configure Maestro Cue - Set up event-driven automation for this agent
  • Copy Agent GUID to Clipboard - Copy the agent’s unique identifier
  • Remove Agent - Delete the agent from Maestro
The git actions here act on the agent you right-clicked, so you can pull or inspect the log of a background agent without switching to it first. Drag the right edge of the sidebar to resize it. The width is persisted across sessions.

​
Collapsed Mode

Click the sidebar toggle (Opt+Cmd+Left / Alt+Ctrl+Left) to collapse the sidebar to icon-only mode. In collapsed mode:
  • Agents show as icons with status indicators
  • Hover for agent name tooltip
  • Click to select an agent

​
Tab Management

Each agent session can have multiple tabs, allowing you to work on different tasks within the same project workspace.

​
Automatic Tab Naming

When you send your first message to a new tab, Maestro automatically generates a descriptive name based on your request. This helps you identify tabs at a glance without manual renaming. How it works:
  1. When you start a new conversation in a tab, your first message is analyzed
  2. An AI generates a concise, relevant name (2-5 words)
  3. The tab name updates automatically once the name is generated
  4. If you’ve already renamed the tab, automatic naming is skipped
Examples of generated tab names: Configuring automatic tab naming:
  • Go to Settings (Cmd+, / Ctrl+,) → General
  • Toggle Automatic Tab Naming on or off
  • Default: Enabled
Automatic tab naming uses the same AI agent as your session, including SSH remote configurations. The naming request runs in parallel with your main prompt, so there’s no delay to your workflow.

​
Manual Tab Renaming

You can always rename tabs manually:
  • Right-click a tab → Rename Tab
  • Or double-click the tab name to edit it directly
  • Manual names take precedence over automatic naming

​
Snoozing Tabs

Snooze hides an AI tab until a moment you choose, then brings it back with a notification you have to dismiss. It’s the email-snooze idea applied to conversations: park work you can’t act on yet without closing it or letting it clutter the tab bar. Hover a tab and choose Snooze Tab, press Opt+Cmd+S / Alt+Ctrl+S, or run Snooze Tab from Quick Actions (Cmd+K / Ctrl+K). Snoozing is available on AI tabs only. Choosing when it comes back The snooze dialog gives you three ways to pick a time, and always previews the exact moment it resolved to before you commit:
  • Presets - Later today, This evening, Tomorrow, This weekend, Next week, Next month. Presets that have already passed for the day drop off the list.
  • Free-form text - type it the way you’d say it:
  • Calendar - pick a date from the month grid and set a time of day.
Note to self Every snooze takes an optional note, and that note becomes the body of the notification when the tab returns. This is what turns snooze into a reminder system: leave yourself the reason you’re coming back (“check if the migration finished”, “review this before standup”) instead of rediscovering it later. What happens while a tab is snoozed The tab disappears from the tab bar and from tab navigation. The conversation is preserved exactly as you left it, and the tab returns to its original position when it wakes. If you snooze an agent’s only tab, a fresh empty tab takes its place so you’re never left staring at an empty workspace. When the time arrives, the tab reappears and Maestro raises a notification that stays until you dismiss it, so a reminder can’t scroll past unseen. Click it to jump straight to the tab.
Wakes are delivered by the running app. If Maestro is closed when a snooze comes due, the tab returns the next time you launch - overdue reminders are never silently dropped.
Long snoozes are safe. Your AI provider owns the conversation transcript and eventually ages old ones out, which would leave a tab snoozed for months coming back empty. Maestro keeps its own copy for the length of every snooze, exactly as it does for starred sessions, and restores it when the tab wakes. That copy is held until the snooze ends, so unstarring a snoozed session does not discard it either. Managing snoozed tabs Open the list from the search icon in the tab bar → See All Snoozed Tabs, or run See All Snoozed Tabs from Quick Actions. It shows every snoozed tab across all agents, soonest first, with its note and a countdown. Each row offers:
  • Unsnooze - bring the tab back right now
  • Reschedule - pick a new time or edit the note
  • Dismiss - drop the snooze and the tab, for when you no longer care
Snooze history Click View History in the Snoozed Tabs header to see snoozes that have already finished. It is one chronological list across every agent, newest first, and each entry keeps the note you left yourself along with when it was due and when it actually came back. Entries are marked by how they ended: came back on schedule, brought back early, or dismissed. Click any entry to jump back to it. If the tab is still open you land directly on it; if only the agent is still around you land there instead. Entries whose agent has since been deleted are shown but not clickable. The log keeps the most recent 100 entries; older ones drop off as new ones arrive.
Dismissing only discards Maestro’s tab. The underlying conversation is still on disk and can be reopened from the Session Explorer.

​
Session Management

Browse, star, rename, and resume past sessions. The Session Explorer (Cmd+Shift+L / Ctrl+Shift+L) shows all conversations for an agent with search, filtering, and quick actions. Session tracking