PaneScale documentation
Everything you need to run several coding agents side by side — safely, verifiably and for longer than you can watch them.
PaneScale is a Windows desktop app for people who use coding agents every day. It starts the command-line agents you already have — Claude Code, Codex, Gemini CLI and seven more — in a grid of real terminals, gives each one its own Git worktree when you want it, and adds what is missing as soon as you run more than one: live status for every session, a dashboard of who needs you, checkpoints, budgets, signed approvals, your subscription limits at a glance and a queue that keeps working while you sleep.
PaneScale doesn’t replace your agents and doesn’t sit between you and your provider. Every pane is the real CLI, signed in with your own account. Everything PaneScale adds — names, status, worktrees, checks, reports — happens on your computer.
What’s in this version
These docs describe PaneScale 0.3. The most important things that are new or changed since the first releases:
| Area | What changed |
|---|---|
| Instant panes | Panes open in milliseconds: ready sessions wait in the background and starts are staggered so the window stays smooth. Code mode |
| Session names | Every session names itself after its first task and shows what it is doing right now. Sessions |
| Dashboard | A live list of all sessions, grouped by urgency, in the right side panel. Dashboard |
| Side panel | Files, search, a project thread, workflows and the browser now open in one resizable panel on the right, chosen from a rail. Side panel |
| Usage & limits | Your Claude and Codex subscription limits, a forecast and your local spend — bottom left and on their own page. Usage & limits |
| Grid up to 30 | Layouts from 1 to 30 panes, every seam and crossing draggable. Code mode |
| Chat | Chats and project threads in the left sidebar, attachments, @ files, prompt enhancement and code review. Chat |
| Tasks board | Cards run on an agent directly, land in review and are approved by you. Tasks board |
How PaneScale is organized
PaneScale has five modes, switched in the title bar or with Ctrl+Alt+1 to Ctrl+Alt+5. Switching never stops anything — terminals keep running and answers keep streaming while you look elsewhere.
| Mode | What it is for |
|---|---|
| Agent | Persistent teammates with a role, rules, an engine, allowed folders and long-term memory |
| Code | The terminal grid — up to 30 panes, worktrees, launch sets, Dispatch |
| Chat | Chats without a project and coding threads inside one, stored locally |
| Workflow | Multi-step runs: agents, model prompts, tests, browser checks, Git review, MCP tools, approvals |
| Video | Product videos written as code on a timeline |
Around the modes sit stages that take over the window for a focused job — the Overnight Autopilot, Usage & limits and Settings — and overlays you can open from anywhere: the command palette, Tasks, Routines, the Workspace hub, notifications and Voice. Tools that belong next to your work — dashboard, files, search, a thread, workflows and the browser — open in the side panel on the right.
The window at a glance walks through every part of the window.
Where to start
- Install PaneScale and sign in to at least one coding CLI.
- Follow Getting started to run two agents side by side.
- Read Sessions & status — it explains what the coloured dots mean and how sessions name themselves.
- Read Lanes & worktrees — it explains why parallel agents don’t step on each other.
- Set budgets and look at your limits before you let agents run longer.
- When you’re ready to let agents work on their own, continue with the Overnight Autopilot and Workflows.
Principles
- Your CLIs, your accounts. PaneScale starts the agents you installed and never sees your passwords or API keys. See Accounts.
- Local first. Database, worktrees, checkpoints, voice and video all stay on your machine. See Your data.
- Evidence over claims. “Ready” means your checks passed on recorded runs — never that an agent said so. See Checks & integrity.
- You decide what leaves your branch. Merges, pushes and pull requests need your click, and risky ones a signed one-time approval.
- Measured, not estimated. Token usage is shown where a CLI reports it and marked “not reported” where it doesn’t. See Token Guardian.