Platforms & requirements

Supported operating systems, the coding CLIs PaneScale works with, the tools some features rely on, and how much hardware many panes need.

Operating systems

PlatformStatus
Windows 11, x64Supported
Windows 10, x64Supported
macOS 11 or later (Apple silicon, Intel)Supported — one universal app, signed and notarized
Linux x86_64 (glibc 2.35+: Ubuntu 22.04+, Debian 12+, Fedora, openSUSE, Arch)Supported — installer, .deb, .rpm

PaneScale is a native app with a Rust core. On Windows it uses the WebView2 runtime that comes with the system. Terminals run on ConPTY, and every pane’s process tree runs in its own Windows job object, so closing a pane always ends everything it started — including MCP servers an agent launched.

PaneScale picks your default shell in this order: PowerShell 7, Windows PowerShell, Command Prompt, Git Bash. Change it in Settings → General; open panes keep theirs.

On macOS PaneScale uses the system WebView (WebKit), on Linux WebKitGTK 4.1. Terminals run on a pseudo-terminal, and closing a pane ends its whole process group. The default shell is your login shell (usually zsh on macOS, bash on Linux); bash, zsh, fish, PowerShell 7 and sh are offered when installed. Agents started from PaneScale get the same PATH as your login shell, so CLIs installed with npm, Homebrew, pnpm, fnm or nvm are found. Keyboard shortcuts use ⌘ instead of Ctrl on macOS.

Coding CLIs

CLIModeNotes
Claude CodeStructured eventsLive activity, hook-based status, session names from its prompts, live limits, effort switchable per pane
CodexStructured eventsLive activity, usage at the end of each turn, status from terminal notifications
Gemini CLITextStatus from terminal signals
GitHub Copilot CLITextGitHub context built in
Cursor AgentText
Grok BuildText
AiderTextSmall steps, commits every change
Kimi CLIText
DroidText
OpenCodeText

Structured CLIs stream machine-readable events, so PaneScale can show each step live and measure token usage exactly. Text CLIs work in every pane; their status comes from terminal signals and screen detection, and their usage is shown as “not reported” instead of being estimated.

Features that run agents non-interactively — Agent mode, Chat, the tasks board, workflow agent steps, routines, the Overnight Autopilot and Point & Fix — use Claude Code or Codex.

Start options per CLI

OptionClaude CodeCodexGemini CLI
Effort--effort, switchable live per panemodel_reasoning_effort—
Bypass--dangerously-skip-permissions--dangerously-bypass-approvals-and-sandbox--yolo

See Accounts → Start options.

Other tools

ToolNeeded for
GitWorktrees, checkpoints, the autopilot, Point & Fix, the review in Chat
gh CLIImporting GitHub issues and opening draft pull requests
OpenSSH clientRemote SSH terminals (built into Windows 10 and 11)
Ollama or your own endpointModel steps in workflows
whisper.cpp (installed from Settings)Voice dictation and commands
Node.js / npmOnly if your project’s checks use npm scripts — PaneScale proposes checks from package.json

Hardware

Every pane runs its own agent process, and agents index your workspace when they start. A waiting Claude Code session with its MCP servers needs roughly 450 MB of memory, a Codex session about 150 MB, a plain shell about 60 MB.

PanesRecommended memory
1–48 GB
6–1216 GB
16–3032 GB or more

Ready sessions adapt to what your computer has: PaneScale always keeps 4 GB of memory free, fills at most four waiting sessions at the same time at lower priority, and starts fewer when memory is tight. The toolbar in Code mode and the footer of the dashboard show how many processes your panes run and how much memory they use.

A display of 1920 × 1080 is enough for up to six panes; for 9 or more, a larger or higher-resolution screen makes terminals easier to read — or use the Compact and Dense grid densities.

Something unclear or missing? Tell us — we improve these pages every week.