glossary

When one AI agent manages other agents for you. It picks up tasks, creates workspaces, starts agents, checks on them, and reports what needs your attention.

A tool that works with any AI agent, not just one. You're not locked into Claude Code, Codex, or any provider. If it runs in a terminal, it works.

A desktop tool that helps you run multiple AI coding agents in parallel. It handles workspace isolation, switching between agents, and git workflow.

When one AI agent coordinates others: breaking work into tasks, delegating to specialized agents, and combining results. Like a tech lead managing a team.

The list of public keys a machine accepts for logins, kept in ~/.ssh/authorized_keys. Add someone's public key here to let them in; remove it to revoke access.

A secret key that proves you're allowed to connect to something. Generated automatically. Treat it like a password.

Command Line Interface. A program you control by typing commands in a terminal. Agents love CLIs because they're already typing commands all day.

The device you're sitting in front of. It shows the UI and sends your input to the host machine where the work actually happens.

A background program that runs quietly on your machine, doing work when asked. Pane's daemon manages your sessions and terminals behind the scenes.

A tool that keeps several terminal shells in one window. You can split panes, switch tabs, and reconnect to a live session later.

Leaving a terminal session while its host keeps running, then connecting to that same live session later.

The host process and environment where an AI coding agent runs, including its terminal, files, credentials, and network access.

The guarantees about what remains after detach, restart, sleep, or a host stop. Saved layout and running processes are different guarantees.

A Pane host running on a cloud VM you create and control. Agents and worktrees run on the VM, and you connect from desktop Pane or your phone. The VM's disk keeps your files when it stops; running processes don't survive a stop.

Using a laptop with its lid closed while it stays connected to power and an external display or input device.

A Progressive Web App, a website that can be installed like an app and connect to a remote host through the browser.

A separate copy of your repository on its own branch. Each worktree shares the same git history but has its own files, so agents can work in parallel.

The machine that does the actual work: runs your code, agents, terminals, and git repos. Your code always stays on the host.

A machine you SSH through to reach another machine that isn't directly reachable. Also called a bastion host.

A network address (127.0.0.1) that only your own machine can reach. When something listens on loopback, nothing outside your machine can connect to it.

A standard that lets AI agents connect to external tools like GitHub, Linear, Jira, and Slack through one shared format.

Mobile Shell. A replacement for an SSH session that keeps working when your connection drops, changes networks or your laptop sleeps.

An agent that coordinates other agents instead of doing the work itself. It decides what needs doing, assigns it, and checks that it got done.

In the Pane app, a pane is an isolated workspace for one feature. Each pane can have its own git worktree, branch, and running agents.

A pane is a workspace (one task, one branch). A panel is a tab inside that workspace (one agent, one tool). Create panes for different features, panels for different tools on the same feature.

A tab inside a pane. Each panel runs its own terminal, agent, or tool. You can have multiple panels in the same pane.

Making a port on one machine reachable from another. Remote Pane does this for you, so a dev server on your remote machine opens at localhost on your laptop.

A virtual terminal that lets programs run and control command-line tools as if they were in a real terminal window.

Writing and running code on a different machine from the one you're sitting at, like a Mac mini, home server or cloud VM, while using it from your laptop or phone.

Pane's self-hosted remote access. One machine (the host) runs your projects and agents; any desktop or phone connects to it with a pane-remote:// code and gets its terminals, files and dev servers.

Secure Shell. The standard way to open a terminal on another computer over the network, encrypted end to end. You type commands on your machine and they run on the other one.

A text file (~/.ssh/config) that gives your machines short names. Write the address, username and key once, then connect with just ssh <name>.

A pair of files that proves who you are without a password. The public key (.pub) goes on machines you want to reach and is safe to share. The private key stays on your computer and is never shared.

Your private Tailscale network: the devices signed in to it, each with a stable address. Other devices and people on the tailnet can connect as far as its access rules allow; the public internet can't.

A tool that puts your devices on one private network, wherever they are, without opening ports on your router. Pane uses it by default to connect your laptop or phone to your other machines.

Each agent gets its own folder, branch, and file copies. They can all work at the same time without overwriting each other.

Windows Subsystem for Linux. It lets you run a full Linux environment inside Windows, so you can use Linux tools without a separate machine.

The program around an AI model that turns it into an agent: tools, a loop, context and permissions. Claude Code, Codex and Pi are harnesses.

A service that runs coding agents like Claude Code or Codex behind one API, in sandboxes, so products can embed them.

A standard for how apps and editors talk to coding agents: JSON-RPC messages over stdin and stdout.

The layer that starts, tracks and steers agents without being an agent itself. It can run locally or as a hosted service.

Passing a task from one agent or session to another with enough context for the new one to carry on.

Folders of instructions, usually a SKILL.md file, that teach a coding agent how to do a task. Loaded only when relevant.

Click an element in a tool's built-in browser and send it to the agent as context: HTML, styles, a screenshot, source location.

A quick aside with your agent in a forked copy of the conversation, so the main task isn't interrupted. Often /btw or /side.

A workspace built around running AI coding agents rather than around you editing code. Close to an agent manager.

A system of AI agents that turns tasks into shipped code, with people approving at a few checkpoints.

A tool that runs several coding agents at once and shows which one needs you, like tmux does for shells.