jumpterm
The whole model in a few minutes

How Jumpterm works

Run your AI coding agents in the cloud or on your machine, and drive them from the web, your phone, or the Mac app.

How it works

A small jumpterm daemon runs on your machine — that’s where the agent and your files live. The web and iOS apps are thin clients that relay to it over an encrypted channel. When your machine is offline the apps tell you, and you can re-launch onto any online host. Your code and keys never live on our servers.

Quick start

1

Install the CLI

One line (needs Node.js 20+), or npm i -g jumpterm. Prefer an app? The Jumpterm Mac app bundles everything (request beta access).$ curl -fsSL https://jumpterm.com/install.sh | sh
2

Sign in

Authenticate this machine — it prints a short code and a link; open it, confirm, done.$ jumpterm login
3

Run your first agent — no API key

Jumpterm finds a coding agent you already have (Claude Code, Codex, or OpenCode) and runs it with its own login, in the current repo. No key to paste.$ jumpterm start "summarize this repo"
4

Host from your phone & web

Leave the daemon running to make this machine an always-on host — then launch and steer sessions on it from jumpterm.com or the iOS app, streaming live to every device.$ jumpterm daemon

Concepts

Host
A machine running the daemon. Sessions run on a host; you choose which one when you launch.
Session
One agent run. Event-sourced, so it's resumable and consistent across every device — refresh or switch devices and you pick up exactly where you were.
Provider account
Your LLM login or BYOK key, per agent. Add a key in Settings and we verify it live; keep several and pick one per session.
Approvals
When an agent wants to run a command or edit a file, it asks. You approve or deny from any device, with the real diff in front of you.
Fleet
All your active agents grouped by project, so you can run many at once and see what's live at a glance.
Jump OS
The local optimizer your sessions run through. It meters real usage, caches and dedupes upstream calls, maximizes provider prompt caching, and enforces per-session token budgets — entirely on your machine, with your key held only in memory. Toggle it per session in the composer; see what it saved in the session view, and your burn rate in Settings → Usage window.

Self-hosted relay

Point the daemon at your own relay before logging in:

$ JUMPTERM_API_URL=https://your-relay… jumpterm login