Work inside a sandbox

Your AI builds the app — but sometimes you need to get your own hands on it. Drop into a shell, watch a run as it happens, or open a full editor in the browser. Three ways in, all scoped to the sandbox's /workspace.

Requires CLI v0.6.4+

agentry sh, agentry logs, and agentry vsc need CLI v0.6.4 or newer. Update with:

curl -fsSL https://agentry.run/install.sh | sh

Check yours with agentry version.

Each command targets the current sandbox when you leave <sandbox> off. Pin one with agentry sandbox use <id>, or name it explicitly.

Pick the right tool

You want to…	Use
Poke around quickly — ls, cat, run a one-off command	agentry sh
Watch a build or server as it runs	agentry logs -f
Make hands-on edits with a real editor + terminal	the in-browser editor

A shell inside the sandbox

agentry sh opens an interactive shell inside the sandbox, landing you in /workspace — the project root.

agentry sh                # the current sandbox
agentry sh ecommerce-store # a specific sandbox

It runs over the broker tunnel, so it works from anywhere your CLI is signed in — no SSH, no IP, no keys. Because it's interactive, run it from a real terminal (not a script or CI step).

Use it for a quick poke: check what files the AI wrote, tail a file, run npm test, inspect package.json, see what's listening. Exit with exit or Ctrl-D.

Project logs

agentry logs shows the sandbox's project logs — what your app is printing as it builds and runs.

agentry logs              # last 100 lines from the current sandbox
agentry logs -f           # stream live, like tail -f
agentry logs -n 500       # last 500 lines
agentry logs my-app -f    # follow a specific sandbox

Flag	Description
-f, --follow	Stream logs live. Ctrl-C to stop.
-n, --lines N	How many lines to show in snapshot mode (default 100).

Reach for agentry logs -f when you've asked the AI to start the dev server or kick off a build and you want to watch it happen — errors, stack traces, and request logs scroll by in real time.

Project logs vs deployment logs

These are the logs from your app running in the sandbox while you build. Logs from a deployed app live in the dashboard — see Logs + status.

The in-browser editor (VS Code)

For real hands-on work, open the in-browser editor (VS Code) — a full editor with an integrated terminal, running in your browser and scoped to the sandbox's /workspace. It's served behind your dashboard login, so only you can reach it.

Open it two ways:

From the dashboard — go to the sandbox's detail page and click Open editor.

From the CLI — agentry vsc opens it in your browser:

agentry vsc               # the current sandbox
agentry vsc landing-page  # a specific sandbox

The first open takes ~15–20 seconds while the editor starts up. After that it's instant.

It's the same VS Code you know: file tree, multi-tab editing, search across the project, and a built-in terminal that's already at /workspace. Use it when you want to edit files directly, run a sequence of commands, or debug something the AI got almost right.

Existing sandbox? Update the server first

The editor needs the sandbox to be on a current runtime image. If Open editor does nothing on an older sandbox, the operator should hit Update server in the dashboard, then try again.

Next

• CLI reference — agentry sh, agentry logs, agentry vsc, and every other command.
• Download a sandbox — agentry pull to get the project onto your machine.
• Logs + status — logs for a deployed app.