# OpenTerm: full feature reference > OpenTerm is a free, keyboard-first terminal organizer for Windows 10 and 11: every terminal in one window, as named sessions per project with splittable panes, built for vibe coding with AI coding agents. Run Claude Code, Codex CLI, Gemini CLI, OpenCode, Cursor Agent, Grok Build, Aider and other agent CLIs side by side, one session per project. When an agent needs you, OpenTerm tells you which one. Everything is saved locally. - Website: https://www.openterm.app - Documentation: https://www.openterm.app/documentation - Download: https://github.com/SurviveANDcraft/OpenTerm/releases - Source code: https://github.com/SurviveANDcraft/OpenTerm - Short index: https://www.openterm.app/llms.txt - Maker: Filip Mercioniu ## At a glance | | | |---|---| | What it is | A terminal organizer and AI coding agent workspace for Windows | | Platforms | Windows 10 and Windows 11. Not available for macOS or Linux yet | | Price | Free. No account, no sign-up | | License | Source-available under the Elastic License 2.0. The code is public on GitHub | | Built with | Tauri 2 (Rust), ConPTY via portable-pty, xterm.js with WebGL rendering | | Shells | Windows PowerShell (default), PowerShell 7, Command Prompt, Git Bash, or a custom shell path | | One-click agents | Claude Code, Codex CLI, Gemini CLI, OpenCode, Cursor Agent, Grok Build | | Recognized agent CLIs | claude, codex, gemini, opencode, cursor-agent, grok, aider, amp, goose, crush, cline (also through npx, pnpm, yarn, bunx, uvx, pipx) | | Data storage | Local only: `%APPDATA%\com.openterm.app\state.json` | | API keys | Encrypted in Windows Credential Manager, never in state.json | | Updates | Signed automatic updates from GitHub Releases | ## Who it is for - Developers who run several AI coding agents at the same time on Windows. - People working across several repositories who want one saved layout per project. - Keyboard-first users who want tmux-style splits and shortcuts in a native Windows app. - Anyone who ends the day with a dozen terminal windows open and wants them organized into one, with plain shells (PowerShell, PowerShell 7, Command Prompt, Git Bash) treated the same as agent panes. - People vibe coding who want to see what Claude Code and Codex are costing, and to dictate prompts instead of typing them. ## 0. First launch - The first time OpenTerm opens it shows a short welcome: pick a theme (applied live), a terminal text size, a three-point tour of sessions, panes and the inbox, and the first project folder plus what each terminal should launch. - It is skippable at every step and appears only once. Replay it from Settings, About & Data, Replay welcome. ## 1. Sessions (one per project) - A session is a named workspace in the left sidebar. It has its own split layout, color, starting folder and task list. - The new-session wizard sets the name, number of terminals, starting folder (with a folder picker or a built-in navigator terminal) and an optional startup command. Presets: claude, codex, opencode, n8n, or a custom command. - Session settings: name, one of 8 colors, and starting directory. - Rename with a double-click or F2. Drag to reorder. Jump to sessions 1-9 with Alt+1-9, or cycle with Ctrl+Tab / Ctrl+Shift+Tab. - Inactive sessions keep running in the background, so switching is instant and scrollback is kept. - Archive a session to hide it and stop its terminals. Its layout, tasks and agent commands are kept, and you can restore it from Settings, Sessions & Backups. - Each session shows a status dot: an agent is working, an agent is waiting for your input, or an agent hit an error. - The session menu has Session settings, Stats (usage), Git Map, Archive and Delete. ## 2. Split terminal panes - Split any pane right or down, as many times as you like, into rows and columns. - Drag a pane's title bar onto another pane: drop on an edge to split there, or on the center to swap the two. - Drag dividers to resize panes, or resize from the keyboard with Ctrl+Alt+Arrows. - Zoom (maximize) one pane with Ctrl+Shift+Z. - Fold a pane down to its title bar with Ctrl+Shift+D. It keeps running. Click the freed space to fill it with a new terminal. - Rename a pane (for example "api" or "worker") so agents are easy to tell apart. - The title bar shows the logo of the agent or shell running there: Claude, Codex, Gemini, OpenCode, Cursor, Grok, PowerShell, cmd, bash, zsh, WSL. - Move focus between panes with Ctrl+Arrows. - Fast terminal rendering with WebGL, clickable links, and copy/paste through the system clipboard. - Adjust font size with Ctrl+= / Ctrl+- / Ctrl+0. ## 3. AI coding agents - The sidebar "+" menu adds a pane that launches PowerShell, Claude Code, Codex CLI, OpenCode, Gemini CLI, Cursor Agent, Grok Build, or a browser. - Agent panes are ordinary terminals, so every terminal feature works with every agent. Any CLI that runs in a terminal works. - Resume after restart: OpenTerm remembers the agent command you typed in each pane and replays it when the app starts again. Claude Code and OpenCode continue their previous conversation. Claude Code, OpenCode and Codex can reopen the exact conversation by session id. You can switch this off to start fresh. - Agent CLI updates: OpenTerm checks the installed and latest versions of Claude Code, Codex CLI, Gemini CLI, OpenCode, Amp (npm) and Aider (PyPI). When one is outdated, an inbox item offers a one-click update, and panes running that agent are refreshed. The status is also in Settings, Updates. ### Agents panel - Ctrl+Shift+A opens a panel on the right that lists every agent you have running as a card: what it is, its session, whether it is working, waiting for you or errored, and the last prompt you gave it. Agents with sub-agents running show how many. Agents that need you sort to the top. Plain shells are listed separately. Filter to one session or all of them. - The panel sits beside the sessions, not over them, so terminals and browser panes narrow while it is open. It is resizable. - Optional AI summaries (OpenRouter key): each card gets a one-line name for its task and current state. They refresh when an agent gets a new prompt or changes state, and nothing is sent while the panel is closed. - Chat: ask questions about your agents in plain language. The assistant can read a terminal's screen, search across terminals, look up an agent's details and check which processes are running and in which terminal. It links the terminals it mentions. Replies stream, and each chat shows its cost. - Modes, picked under the chat box: Ask (read-only, changes nothing), Act (can also open, split, resize, focus, rename and close panes, run a command in an idle shell and prompt an idle agent; every action waits for your approval on a card in the chat) and Auto (acts without asking, except for risky actions). Act and Auto are off until enabled in Settings, AI. An optional chime plays when an approval is waiting. - Managing the app: in Act and Auto the assistant can also create and edit tasks, create and edit sessions, show or hide the sidebar and panels, and change appearance, terminal and notification settings. It can report usage and cost per terminal or session in any mode. Task edits run without asking; every setting change needs approval, even in Auto. API keys, the assistant's own modes, model and permissions, the shell, dictation and keybinds are off limits to it. - Permissions: Settings, AI has a switch for each area the assistant can reach (panes, typing into terminals, tasks, sessions, panels, usage, settings). - Reporting back: commands it runs and prompts it sends are watched; when the terminal finishes, the assistant is notified and tells you the result. It can also watch any other terminal on request. - Built-in limits in every mode: it never answers an agent's own permission prompt, only types into an idle shell or an agent that is not mid-task, and only acts in the sessions in scope. Deleting files, force-pushing, installing packages, piping downloads into a shell, touching credentials, chaining commands and closing a busy pane always need approval, even in Auto. Stop cancels everything queued. - Model: DeepSeek V4 Flash by default; any OpenRouter model can be set. The assistant can make mistakes in every mode, so check what it reports and review actions before approving them. - Drag a terminal's title bar onto the panel to attach it to the chat and ask about just that terminal. - Attach files: drop files on the panel (from the Files sidebar or Windows) to attach them to the next question. The assistant reads their content: text files as they are, Word documents as text and spreadsheets as CSV, up to 12,000 characters per file. PDFs, images and media go along as a path only. ### Voice assistant - The waveform button in the Agents panel chat starts a live spoken conversation with the same assistant. It takes over the panel while it runs; ending it moves everything said into the chat thread so you can carry on typing. - It follows the selected mode (Ask, Act or Auto) with the same permissions, approval cards and watches as the typed chat. Talk over it to interrupt. M mutes the microphone. - Eight Gemini voices with a preview in Settings. Avatars: an orb (default), Buddy (a blob) or Byte (a tiny retro monitor), which lip-sync, follow the cursor, doze off when muted and react when poked. - Conversations outlast a single connection: OpenTerm renews it in the background and keeps the context, and reconnects on its own after a drop. - The dictation vocabulary is passed to the assistant so project words are heard correctly. - Runs on Google's Gemini Live with your own Gemini API key (free in Google AI Studio). While a conversation runs, microphone audio and the terminal text the assistant reads go to Google; on Google's free tier they may be used to improve Google's products. The key itself never reaches the app's web view: voice connects with a single-use token. ## 4. Inbox and attention alerts - The bell in the title bar (Ctrl+Shift+I) collects events from sessions you are not watching. Each item names its session and takes you to the pane or task. - Item types: approval prompts, errors, finished commands, exited processes, due tasks, delegation updates, backup notices, app updates, repos behind GitHub, and agent CLI updates. - Detects real interactive prompts, such as y/n questions, "press enter to continue", and the numbered menus agents use to ask for approval. It ignores ordinary agent chatter. - A pane that is waiting for input gets an indicator on its title bar. - Optional AI naming: with an OpenRouter key, raw terminal output (box-drawing characters, stack traces) becomes a short title and a one-sentence summary. The original text is still shown on hover. - Sound alerts with one of seven built-in chimes (Classic, Glass, Mallet, Halo, and the short Dot, Duo and Pluck) or your own sound file, plus a taskbar icon flash. Each can be turned off. - Each notification type can be turned off on its own. ## 5. Command queue - Press Ctrl+Shift+Enter in a pane to queue a command. It is typed in automatically as soon as the pane is idle at its prompt again. - Works with PowerShell, cmd, bash and agent CLIs like Claude Code and OpenCode. It learns each pane's prompt from your own typing. - Queue several commands in order, for example: run tests, then lint, then git push. ## 6. Task board and delegation to agents - Each session has its own task list (Ctrl+Shift+K), shown as a list or a kanban board, with a filter for this session or all sessions and a search box. - Statuses: To do, In progress, Done. Priorities: Low, Medium, High. - Each task can have a description, subtasks with checkboxes, tags, a due date (due tasks show up in the inbox) and attached local files. - Drag a task off the Tasks panel onto any terminal to paste it in as a prompt with its description, subtasks and attached file paths. Enter is not pressed, so you can add to it first. Esc cancels the drag. - Delegate a task to an AI agent: choose Claude Code, Codex CLI or OpenCode, then choose when it starts: - Start now. - At a set time. - When another terminal finishes its current work. - OpenTerm opens a pane for the agent and types the task in as a prompt, including the description, subtasks and attached file paths. Delegation status is shown on the task and in the inbox. You can cancel a delegation. - Optional: with an OpenRouter key, an AI clarifies the prompt before handing it off. ### GitHub issues - The Tasks panel has an Issues tab listing the GitHub issues of the repo the session's folder belongs to: open, closed, or assigned to you, with search. - Sign-in goes through the GitHub CLI (`gh auth login`, once). The token stays with `gh`; OpenTerm never sees or stores it. - Read an issue and its comments, comment, create an issue, close it as completed or not planned, reopen it, and assign or unassign yourself. - Turn an issue into a linked local task, or delegate it to an agent directly. The agent is asked to put "Fixes #N" in its commit so the issue closes when the work merges. ## 7. AI usage limits and cost tracking - Rate limits: panes running Claude Code or Codex show how much of the 5-hour usage window is left and when it resets. They refresh every 30 seconds. - Per-terminal usage and costs (pane menu, "Usage & Costs"): tokens split into input, output, cache write, cache read and reasoning, with estimated USD cost by model and by agent session. - Whole-folder stats: usage for every agent session run in a project folder. - Supported: Claude Code (matched exactly), OpenCode and Codex (matched by timing and folder). - Usage history stays with a pane after the agent exits, and across restarts. ### Speech-to-text dictation - Push-to-talk: hold (or toggle) a global shortcut (default Ctrl+Win), speak, and the text is typed where your cursor is. Transcription uses Microsoft MAI Transcribe via your OpenRouter key. - Customizable: shortcut, hold or toggle mode, pill or glow indicator and its screen position, start sound, microphone, language (or auto-detect), clean or verbatim style, custom vocabulary (also used by the voice assistant), maximum recording length, and always-on or only while OpenTerm is open. - Optional AI polish runs the text through DeepSeek for punctuation and cleanup. - Slow answers: if transcription is slow to come back, a second request runs beside the first and the faster answer wins. - Cancelling: Esc cancels while recording. Once the text is on its way, press Esc twice; the recording is kept under Unsent to send again later. - In an OpenTerm terminal, dictation pastes with the app's own paste shortcut, so it keeps working when that shortcut is rebound. ## 8. Embedded browser panes - A real Chromium (WebView2) browser in a pane next to your terminals, with an address bar that also searches. Useful for previewing a dev server while an agent edits code. - Console and network capture: the pane records the page's console messages, JavaScript errors and fetch/XHR requests. - Share page context with an AI agent: Alt+drag a browser pane onto a terminal to paste its console and network log there. Alt+Shift+drag sends a screenshot. You can also copy the log to the clipboard. - Device emulation with real viewport size, pixel ratio, user agent and touch, drawn inside a device frame, plus portrait/landscape rotation and a free responsive mode. - 33 device presets: - iPhone Duo (folded and unfolded), iPhone 16 Pro Max, 16 Pro, 16 Plus, 16, 15 / 14 Pro, 14 / 13, 13 mini, SE (3rd gen) - Pixel 9 Pro, Pixel 8, Pixel 7a - Galaxy S24 Ultra, S24, A54 - Galaxy Z Fold 6 and Fold 5 (cover and unfolded), Galaxy Z Flip 6, Pixel Fold - iPad Pro 13", iPad Air 11", iPad (10th gen), iPad mini, Galaxy Tab S9, Surface Pro 9 - Laptop 13", Laptop 15", Desktop 1080p, Desktop 1440p - Foldables (iPhone Duo, Galaxy Z Fold) have a fold button that switches between the folded and unfolded screen with a hinge animation. Switching between devices that share a browser keeps the page state. - Browser panes remember their URL, device and orientation across restarts. ## 9. Files: explorer, editor and previews - The sidebar Files tab lists the active session's folder. Filter files and refresh the list. - Right-click a file for Open in Explorer, Copy Path, Rename and Delete. - Drag a file from the sidebar, from Windows Explorer or from the desktop onto a terminal to type its path, quoted when needed. - Code editor (CodeMirror) with syntax highlighting, autocomplete, bracket matching, code folding, rectangular selection, undo history and search. Save with Ctrl+S. Supports JavaScript, TypeScript, Python, Rust, C, C++, Java, PHP, SQL, HTML, CSS, JSON, YAML, XML, Markdown and more. - Visual Markdown editor with a toolbar (bold, italic, strikethrough, code, headings, lists, quotes, links, rules, tables). A file opens in visual mode only if it passes a round-trip check unchanged, so your Markdown is never quietly rewritten. - Read-only previews: - Images: PNG, JPG, GIF, WebP, BMP, ICO, AVIF, SVG - Video: MP4, WebM, MOV, MKV, AVI and more - Audio: MP3, WAV, OGG, FLAC, M4A, AAC, Opus - Documents: PDF, Word (.docx) - Spreadsheets: Excel (.xlsx, .xlsm, .xlsb, .xls), OpenDocument (.ods), CSV, TSV ## 10. Git and GitHub - GitHub freshness check: OpenTerm runs a background `git fetch` for each session's repo and tells you in the inbox when your branch is behind GitHub. It also flags local commits and uncommitted changes. It never pulls, merges or resets on its own. - "Review with AI": from that inbox item, open Claude Code, Codex or OpenCode in the session with a ready-made prompt to review the sync state. - Git Map (session menu): a read-only view of the repository. - Explorer tree: working tree, branches, remotes, tags, stashes, contributors. Nested branch names become folders. - A commit graph with branch lanes, a commit filter, zoom, and copy-sha. - Ahead/behind counts against upstream, and commits per week for the last 12 weeks. - Uses local git data only. ## 11. Search - Find in terminal (Ctrl+Shift+F): step through matches with Enter / Shift+Enter. - Highlight search (Ctrl+S): highlights every match in the focused pane and keeps them lit while you scroll, until Esc. ## 12. Windows integration - Open a folder from File Explorer: type `OpenTerm` in the address bar (the same way `cmd` works), or right-click a folder or empty space in it and choose "Open in OpenTerm". - If OpenTerm is closed, it starts with a session for that folder. - If it is open, a new session for that folder is added to the same window. - If that folder already has a session, OpenTerm switches to it instead of making a duplicate. - This uses current-user registry keys, so it needs no admin rights. Turn it off in Settings, Windows Explorer, which removes the keys. - Dock external terminals: drag a terminal window from outside OpenTerm into a pane to dock it there. This can be switched off. - Taskbar icon flash when a pane needs attention. ## 13. Backups, restore and undo - Every save also creates a backup in `%APPDATA%\com.openterm.app\backups`: the last 30 changes plus one snapshot per day, kept for 7 days. Saves made close together count as one backup. - If `state.json` is missing or corrupt, OpenTerm restores the newest valid backup and tells you in the inbox. - Backup timeline (Settings, Sessions & Backups): see each backup's time, sessions and pane count, search the list, preview what a restore would bring back or remove, and restore in one click. Your current state is saved first, so you can undo a restore. - Find in backups (optional, with an OpenRouter key): describe what you are looking for, for example "the session where I was rewriting the checkout flow". Only timestamps and session names are sent, never terminal content. - Undo close: closing a session, pane or task shows an Undo toast. The last 3 closed items can be restored for 10 minutes with Ctrl+Shift+U. ## 14. Themes and appearance - 22 themes in three groups (Minimal, Classic, Wild): OpenTerm (default), Paper, Slate, Mono, Ice, Nord, Solarized Dark, Solarized Light, Gruvbox Dark, Dracula, One Dark, Tokyo Night, Monokai, Rosé Pine, Ocean Deep, Forest, Cyberpunk Neon, Synthwave, Matrix, Galaxy, Blood Moon, Sunset. - Themes apply to the whole app, including terminals and the code editor. - Settings: font size, terminal font (Cascadia Mono by default), cursor style (bar, block or underline), cursor blink, pane padding, scrollback lines (8,000 by default), and sidebar width. Toggle the sidebar with Ctrl+B. ## 15. Keyboard shortcuts Every action has a shortcut and every shortcut can be changed in Settings, Keybinds, which warns about conflicts. Shortcuts follow physical keys, so they also work on QWERTZ, AZERTY and other layouts. The cheat sheet opens with Ctrl+/. | Action | Default shortcut | |---|---| | New session | Ctrl+Shift+T | | Close session | Ctrl+Shift+Q | | Rename session | F2 | | Next / previous session | Ctrl+Tab / Ctrl+Shift+Tab | | Jump to session 1-9 | Alt+1 ... Alt+9 | | New terminal in session | Ctrl+T | | Split right | Ctrl+Shift+E | | Split down | Ctrl+Shift+O | | Close pane | Ctrl+Shift+W | | Focus pane left / right / up / down | Ctrl+Arrow keys | | Resize pane | Ctrl+Alt+Arrow keys | | Zoom pane (toggle) | Ctrl+Shift+Z | | Fold pane (toggle) | Ctrl+Shift+D | | Toggle sidebar | Ctrl+B | | Switch sidebar view (Sessions / Files) | Alt+` | | Toggle Agents panel | Ctrl+Shift+A | | Font size bigger / smaller / reset | Ctrl+= / Ctrl+- / Ctrl+0 | | Find in terminal | Ctrl+Shift+F | | Highlight search | Ctrl+S | | Queue command (runs when the current one finishes) | Ctrl+Shift+Enter | | Open tasks | Ctrl+Shift+K | | Open inbox | Ctrl+Shift+I | | Undo last close (session, pane or task) | Ctrl+Shift+U | | Paste (bracketed-paste aware) | Ctrl+Shift+V | | Shortcut cheat sheet | Ctrl+/ | | Open settings | Ctrl+, | ## 15b. Mouse and drag gestures These are fixed (not rebindable) and are not in the keyboard list above. Panes: - Drag a pane's title bar onto another pane: drop on an edge to split there, drop on the middle to swap the two panes. - Alt + drag a terminal's title bar onto another pane: hands that terminal over as context, typing its recent output into the target pane's stdin. Only terminal panes can receive it, because it needs a writable stdin. Alt can be pressed or released mid-drag; the state when you release decides. - Alt + drag a browser pane onto a terminal: pastes the page's console messages, JavaScript errors and network requests in as text. - Alt + Shift + drag a browser pane onto a terminal: sends a screenshot of the page instead of the log. - Alt + click a split button (or hold Alt with the split shortcuts): duplicates the pane you split from instead of opening a blank shell. Also works on the "New terminal" button in the space a folded pane leaves behind. - Double-click a pane title bar: folds the pane to its title bar, or unfolds it. Terminal, browser and file panes. - Drag a divider: resizes the panes either side of it. - Drag a terminal window from outside OpenTerm into a pane: docks it there. - Right-click a pane title bar: opens the pane menu. Files and sessions: - Drag a file onto a terminal: types its full path in, quoted when needed. Works from the Files sidebar, from Windows Explorer and from the desktop. - Ctrl + Alt + drag a file onto a pane: docks the file itself as a pane instead of typing its path, with the drop edge deciding where it lands. Folders ignore this. Alt alone stays path mode on purpose. - Alt + drag a file pane's title bar: expands that file to full screen. Without Alt it moves around the grid like any other pane. - Alt + drag the full-screen file viewer's title bar into the grid: docks it back as a pane. Holding Alt marks the bar as grabbable first. - Alt + drag a session row from the sidebar onto a terminal: types that session's folder path in. Without Alt, dragging a session row reorders the sidebar. - Right-click a file in the Files sidebar: Open in Explorer, Copy Path, Rename, Delete. Inside a terminal: - Ctrl+C: copies when there is a selection; with nothing selected it stays a plain interrupt. - Alt + click a link in terminal output: opens the URL in a browser pane beside that terminal. A plain click uses your default browser. - Ctrl + scroll in Git Map: zooms the commit graph. ## 16. Privacy and security - No account, no sign-up, no telemetry. - Sessions, layouts, tasks, settings and keybinds are stored locally in `%APPDATA%\com.openterm.app\state.json`. - Terminal content only leaves your machine for the optional AI features, through your own keys. - OpenTerm only uses the network for: checking for app updates (GitHub), checking agent CLI versions (npm and PyPI), `git fetch` for the GitHub freshness check, GitHub issues through your own `gh` CLI, pages you open in browser panes, and the optional AI features (OpenRouter, and Google for voice). - Optional AI features (the Agents panel assistant and card summaries, notification titles, prompt clarification, find in backups, dictation) run through your own OpenRouter API key. The voice assistant runs on Google's Gemini Live with your own Gemini API key. With no key set, nothing is sent. - API keys are stored encrypted in Windows Credential Manager, tied to your Windows account, and never written to state.json or its backups. Settings shows only the first and last few characters. Keys are used by OpenTerm's native side alone; voice connects with a single-use token instead of the key. - Updates are signed. The app refuses any update it cannot verify against the maintainer's public key. - Agent CLI updates can only run commands from a fixed list built into the app. ## 17. Settings sections Appearance (theme, terminal font and size, cursor, padding), Terminal (shell, scrollback, docking external terminals), AI & Agents (resume conversations, OpenRouter key, AI inbox names, AI agent card summaries, the assistant's modes, permissions and model, the voice assistant's Gemini key, voice, avatar and model, agent CLI versions), Dictation, Notifications (seven built-in chimes or a custom sound, taskbar flash, per-type toggles), Keyboard (every shortcut, rebindable), Sessions & Backups (archived sessions, backup timeline), About & Data (version and updates, state file location, replay welcome, reset). ## FAQ ### What is OpenTerm? A free Windows desktop app for running many terminals and AI coding agents at once. You get named sessions (one per project), split panes, a shortcut for every action, and one inbox that tells you when an agent needs you. ### Is OpenTerm free? Yes. It is free with no account. The source code is public on GitHub under the Elastic License 2.0. ### Which AI coding agents does OpenTerm work with? Any agent that runs in a terminal. Claude Code, Codex CLI, Gemini CLI, OpenCode, Cursor Agent and Grok Build open in one click. Aider, Amp, Goose, Crush and Cline are also recognized so they relaunch after a restart. ### Does OpenTerm run on macOS or Linux? Not yet. It supports Windows 10 and Windows 11. ### Does my code leave my machine? No. Everything is stored locally, and OpenTerm has no telemetry. The only exception is the optional AI features, which you turn on by adding your own OpenRouter API key (or a Gemini API key for the voice assistant). ### Where are my API keys stored? Encrypted in Windows Credential Manager, tied to your Windows account. They are never written to OpenTerm's settings file or its backups, and Settings only shows the first and last few characters. ### Can I talk to the assistant instead of typing? Yes. The waveform button in the Agents panel starts a live voice conversation with the same assistant, on Google's Gemini Live with your own Gemini API key. For typing prompts into any app by voice, use dictation. ### Is OpenTerm a tmux alternative for Windows? It covers the same ground on Windows: named sessions, split panes, keyboard navigation and saved layouts, all in a native app with no WSL needed. It adds features for AI agents, such as the inbox, command queue, task delegation and usage limits. ### Can I run several Claude Code or Codex sessions at the same time? Yes. Open as many agent panes as you want in one session or across sessions. Each pane shows its agent's logo and rate-limit usage, and the inbox tells you which one is waiting for approval. ### How do I see my Claude Code or Codex usage limit? Panes running Claude Code or Codex show how much of the 5-hour window is left and when it resets. "Usage & Costs" in the pane menu shows token and cost details. ### How do I get notified when an AI agent needs approval? OpenTerm watches every pane for approval prompts and errors. It adds an inbox item, plays a sound, flashes the taskbar icon and marks the session in the sidebar. ### Which shells are supported? Windows PowerShell (default), PowerShell 7, Command Prompt, Git Bash, or any shell executable you choose. ### How does OpenTerm update? It checks GitHub Releases at startup and installs signed updates after you confirm. ### How do I install OpenTerm? Download the latest Windows installer from https://github.com/SurviveANDcraft/OpenTerm/releases and run it.