Web workspace · available now

What will you create?

Describe the task. AgentXPro reads your code, runs commands, edits files and reports back, with every step on the record.

Add rate limiting to /login

ChatTrajectory

Session log

Add rate limiting to POST /login and cover it with a test.

Deep diving…Thought for 6s

Search"login" in src/3 matches

Readsrc/routes/auth.ts142 lines

Editsrc/routes/auth.ts+14 −2

Writetest/login.rate-limit.test.ts+32

Runnpm test18 passing

POST /login now allows 5 attempts per minute per IP. The 6th gets a 429 with a Retry-After header.

router.post("/login", rateLimit({ window: "1m", max: 5 }), login);

The new test covers the blocked attempt and the reset after the window. All 18 tests pass.

Message the agent

Workspace Write

Web available now
Terminal in development
macOS in development
iOS & Android in development

What it does

Does the work. Shows the work.

Every read, command and edit appears as it happens, so you can follow along, step in or approve.

Build in chat. See it live.

Open Preview beside your conversation. Saved app changes appear automatically as the agent edits your HTML, CSS and JavaScript. Resize the pane, switch devices, or pause live updates while you inspect the app. Standalone Focus, Split and Gallery views are available too. Use /preview in the terminal client for a browser preview that follows local file changes.

Open workspace previews ↗

◫ Chat + previewBuild side by side

● Live updatesFollow saved app changes

▦ Device viewsDesktop · Tablet · Mobile

Watch it think, then act

Reasoning streams live next to the answer. Shell commands, file reads, edits and searches each show up as a step you can inspect.

Sandboxed by default

Commands run inside a sandbox rooted at your workspace. Anything outside it asks first.

Run a command outside your workspace?

rm -rf ../shared/cache

DenyAllow once

Isolated workspaces

Each user gets their own agent process and their own storage.

Files & images

Mention files with @ or attach screenshots. It works from what you give it.

Pick up where you left off

Every chat is saved to its workspace. Reopen it days later and keep going.

Install

Bring the agent to your terminal.

One command on macOS, Linux or Windows. Same Google sign-in as the web workspace.

Available now

Terminal

A full-screen client for your shell, with streaming reasoning, slash commands, model switching, permission prompts and a live status line.

curl -fsSL https://agentxpro-site.vercel.app/install.sh | sh

Run in Terminal. macOS 12+ (Apple silicon or Intel) or Linux.

irm https://agentxpro-site.vercel.app/install.ps1 | iex

Run in PowerShell. Windows 10 (1803+) or 11, x64 or ARM64.

No admin rights. Everything installs to ~/.agentxpro. Then run agentxpro and sign in with Google. install.sh · install.ps1 · SHA-256

~/code/api — agentxpro

 █████╗  ██████╗ ███████╗███╗   ██╗████████╗██╗  ██╗██████╗ ██████╗  ██████╗
██╔══██╗██╔════╝ ██╔════╝████╗  ██║╚══██╔══╝╚██╗██╔╝██╔══██╗██╔══██╗██╔═══██╗
███████║██║  ███╗█████╗  ██╔██╗ ██║   ██║    ╚███╔╝ ██████╔╝██████╔╝██║   ██║
██╔══██║██║   ██║██╔══╝  ██║╚██╗██║   ██║    ██╔██╗ ██╔═══╝ ██╔══██╗██║   ██║
██║  ██║╚██████╔╝███████╗██║ ╚████║   ██║   ██╔╝ ██╗██║     ██║  ██║╚██████╔╝
╚═╝  ╚═╝ ╚═════╝ ╚══════╝╚═╝  ╚═══╝   ╚═╝   ╚═╝  ╚═╝╚═╝     ╚═╝  ╚═╝ ╚═════╝

› /health returns 500 when Redis is down. Make it degrade instead.
◆ thinking — reading the handler and its tests…
read src/routes/health.ts
bash npm test -- health ✗ 1 failing
edit src/routes/health.ts +9 −3
bash npm test -- health ✓ 6 passing
✓ /health now returns 200 with {"redis":"down"} and logs the error. Tests pass.

Coming soon

Next: your desktop and your phone.

In development. Each one connects to the same workspace, so a conversation started in one place shows up in the others.

In development

macOS

Native menus, keyboard shortcuts, file panels and sign-in through your system browser.

AgentXPro

Settings page load time

Why does the settings page take 3 seconds to load?

Deep diving…Thought for 4s

Search"fetch(" in src/pages/3 matches

Readsrc/pages/settings.tsx210 lines

Editsrc/pages/settings.tsx+6 −11

Runnpm test42 passing

It fetched /api/user three times on mount, once per panel, one after another. I moved it into a single shared query.

const { data: user } = useQuery(["user"], fetchUser);

The page now makes one request instead of three. All 42 tests pass.

Message the agent

Workspace Write

⇧⌘ONew chat

In development

iOS & Android

Check in on your agent from anywhere.

9:41

Last 3 commits

What changed in the last 3 commits?

Deep diving…Thought for 3s

Rungit log -3 --stat3 commits

ReadCHANGELOG.md48 lines

Three commits since yesterday: rate limiting on POST /login, a fix for the flaky upload test, and a test runner upgrade.

Nothing needs a migration or a config change.

Message the agent

qwertyuiop

asdfghjkl

⇧zxcvbnm⌫

123spacereturn

How it works

Sign in. Ask. Review.

In the browser there's nothing to install. Your private workspace is created the first time you sign in.

1
Sign in
Use your Google account. Your private workspace is created on first sign-in.
2
Describe the task
Fix a bug, write a test, explain a module, refactor a file. Attach files or screenshots if they help.
3
Review what it did
Every command and edit is on the record. Approve, push back or keep going in the same thread.

Security

Your code stays in your workspace.

AgentXPro is built so that one user's agent can never see another's files.

Sign-in: Google sign-in. Every request to the workspace is authenticated.
Isolation: One agent process per user, each with its own state and storage.
Files: Workspace operations are clamped to your directory and can't reach outside it.
Commands: Shell commands run in a sandbox and ask before going further.
Apps: The upcoming native apps sign in through your system browser with a single-use code you approve. No tokens in URLs.