Cloud dev with a2a dev

a2a dev runs your agent in the cloud while you edit locally. It syncs your project to the agent's scale-to-zero dev box, runs the agent's hot-reloading server there, and serves it on a public URL. Save a file and the change is live in seconds — on real cloud infrastructure, not your laptop.

cd my-agent/
a2a dev

You get a URL like https://my-agent-devbox.a2acloud.io/ — a live, shareable endpoint for the agent — and a file watcher that pushes every save up to the box, which reloads on the change.

Why run dev in the cloud

The inner loop stays local — your editor, your keystrokes — but the agent runs where it deploys:

• Parity — the same base image, the same runtime as production. No "works on my machine": if it runs in a2a dev, it runs deployed.
• A public URL for free — point a webhook at it, share it in a PR, demo it to a teammate, or connect it to a client. No tunneling tool to run.
• Nothing to install locally — no Docker, no per-agent Python environment. The box already has the toolchain.

The loop

a2a dev
 → wakes the agent's dev box (cold start takes a few seconds)
 → syncs your project to the box
 → runs the agent's dev server there (hot reload)
 → prints https://<agent>-devbox.a2acloud.io/   ← public, shareable
 → watches local files; every save syncs up and the box reloads

Edit agent.py, hit save, and within a couple of seconds the running agent on the box has your change. The box's dev server watches the synced files, so the reload is automatic — the CLI only keeps the files in sync.

Run on your machine instead

When you're offline, air-gapped, or just want the agent in your own process, use --local:

a2a dev --local              # run on this machine (the classic local dev server)
a2a dev --local --host-runtime  # run directly in your Python, no container

a2a dev --local is the old behavior: it serves the agent at http://127.0.0.1:8000 with hot reload, using .env.local and a local workspace.

How it works

The dev box exposes a single public origin over the platform's HTTPS ingress. A small bridge on the box fronts two things on that origin:

• /ssh — the WebSocket tunnel that carries file sync and shell access (the same channel a2a ssh uses).
• everything else — reverse-proxied to the agent's dev server running inside the box.

So https://<agent>-devbox.a2acloud.io/ reaches your running agent, while the CLI syncs files over the same ingress. No extra ports, no bastion, no firewall rules. Files sync with rsync over the tunnel; the box's own dev server handles hot reload, so a save becomes a reload without any push step of its own.

Requirements

• You must be logged in (a2a login) — the box runs under your account, with your quota and your CLI credentials injected, exactly like a2a ssh. If you're not logged in, a2a dev will tell you; use --local to run without an account.
• rsync on your machine (standard on macOS and Linux).

Reference

a2a dev                      # cloud: sync + run + public URL (inside an agent repo)
a2a dev --agent <name>       # cloud dev for a named agent
a2a dev --local              # run on this machine with hot reload
a2a dev --local --host-runtime  # local, directly in this Python process

The first a2a dev after an idle period cold-starts the box (a few seconds); the CLI waits through it. See also SSH dev boxes and Agents.