FAQ

An experience for directing agents, not for editing files. It gives you one prompt where you say what you want, across all your projects, and agents do all of the work. Instead of opening a window per project and typing commands yourself, you name the project or workspace and the request goes there. Results, questions, and finished work come back to the same prompt. It runs on macOS, Linux, and Windows.

People who direct agents more than they edit files. If you spend your day editing files by hand, an editor is the better tool. If you spend it starting, steering, and reviewing agents across several projects, it’s built for you.

Because it’s the only way to never decide where to type. It’s the core design constraint: there is exactly one prompt, and every feature works through it. The prompt always shows where your next request goes. See The constraint.

No. You type intent, and agents turn it into commands. Every command an agent runs shows up as a block you can read, copy, or hand back with “fix this”. When you want a raw shell, any other terminal is right there.

A terminal was originally the human’s end of a line to a computer doing work somewhere else. New Terminal goes back to that meaning. The machines doing the work are now agents, on your laptop, your servers, and the cloud, and the prompt is your end of the line to all of them.

Yes, for iOS and Android, with the same prompt. Because you only ever type intent, nothing is lost on a phone: give requests, answer agents’ questions, approve permissions, review diffs, and watch any agent’s session.

Any agent that runs in a terminal, including Claude Code, Codex, Cursor, and OpenCode. Hosted agents from providers’ clouds connect through adapters, and the adapter spec is open. A built-in agent ships with the app, so the prompt works the moment you open it, and you can choose any other agent as your default.

Any. Agents that let you pick a model can use any provider’s API key or a local model. Your keys go straight to the provider, never through the app.

By default an agent can only touch the panes it created and its own workspace. Anything more needs your approval. Agent control is always visible, your requests always take priority, one key stops every agent, and every action is logged. See Safety.

A project is something you work on, such as a codebase or a server, and it lasts. A workspace is a task in progress: it brings in the projects it needs and holds the task’s agents and history. Each workspace with a repository gets its own branch and isolated copy, so parallel agents never step on each other.

Yes. Agents can run on your machine, in a local sandbox, on your own remote machine over SSH, in an agent provider’s cloud, or in New Terminal’s hosted cloud. They all look the same: a pane, a status, a workspace, and a diff. Only the hosted cloud is paid.

Warp is the closest thing to New Terminal and a big inspiration. The differences:

• Intent only. You only give intent, and agents do all of the work.
• One prompt. Exactly one prompt instead of an input per tab and pane.
• Mention instead of navigate. You mention a project instead of going to where it lives.
• The keyboard never leaves. Agents drive programs while you watch.
• A terminal, not a platform. One thing: the place where you direct agents.

tmux multiplies places to type: sessions, windows, and panes, each with its own shell. Here you have one prompt. Panes show what agents are doing, and you never switch between them to find your work; you name the project or workspace and the request goes there.

Ghostty is a fast, correct terminal for running your shell, and a big influence. New Terminal works one layer up: you don’t type commands, agents do. You give intent at one prompt across every project, and it shows what agents are doing as blocks and panes.

cmux showed that agents deserve their own panes, status, and notifications. Conductor showed that the workspace is the right unit of work. New Terminal builds on both with one prompt for every workspace, an attention queue, diff review in the terminal, support for macOS, Linux, and Windows, and an open interface that lets agents drive the terminal directly.

Yes, under the Apache 2.0 license. That includes the whole local experience: the app itself, agent hosting, the attention queue, workspaces, diff review, and the built-in agent. Paid features are ones that cost money to run or only make sense for teams. The test: if a solo developer on a plane would miss it, it’s free.

No. An account is only needed for paid features.