Xervente
For developers who use coding agents

Your coding agent keeps working. You don't have to watch it.

Coding agents stop to ask you things all day. Xervente sits between you and the agent: it answers the way you would, and holds anything risky until you say yes, from your desk or your phone.

Join the private beta See how it works
Runs on your computer. Uses the AI you already pay for. Nothing goes to our servers.
Claude CodeCursorGemini CLIAntigravityWindsurf: watch & hold
On the laptop the agent asks: Postgres or SQLite? Xervente answers: Postgres, you use it in every project. You are away, with your phone.
How it works

Three things, all day long

Xervente plugs into your agent's own hooks. There is nothing new to learn: keep working the way you do.

The agent asks: should I add tests first? And waits.
1

The agent stops to ask

"Should I add tests first?" "Postgres or SQLite?" Normally it waits until you come back. That is the time Xervente gives you back.

Xervente answers from what it knows about you: yes, tests first.
2

Xervente answers as you

It learns how you decide from your own sessions and corrections, and replies the way you would, with your model. It also turns down extras you never asked for.

npm publish and an API key are held; your phone asks: approve once or deny.
3

Risky steps wait for you

Publishing, deploying, deleting data, payments, passwords and keys are never decided for you. They come to your phone: approve once, or deny.

Two ways to work

Autopilot, or check every answer first

Pick per project, change any time with /xervente review or /xervente autopilot.

A

Autopilot

Xervente answers the agent itself and only comes to you when it is serious. Your agent keeps going while you are in a meeting, at lunch, or asleep.

R

Review

Xervente drafts the answer. You see what the agent asked and the draft side by side: send it, edit it, or drop it. The agent waits up to ten minutes.

On the sofa, the phone shows what the agent asked and Xervente's draft answer, with Send and Edit.
Always yours

What Xervente never decides for you

These always wait for you, in autopilot too. A password typed on your phone goes to the agent once and is never stored.

Publishing packages and apps
Deploying to production
Deleting data and force-pushing
Payments and anything with money
Passwords, keys and 2FA codes
Your IDE's own permission prompts
Private by design

Everything stays on your computer

Xervente is not a cloud service. It runs next to your agent, on your machine.

  • Your code, your sessions and what Xervente learns about you stay on your computer.
  • Answers come from the AI you already use, with your own account. No second model, no extra key.
  • Your licence is checked offline. Nothing is sent to Arad Soft.
  • Your phone talks to your own computer, paired with a QR code.
Your computer at home, locked; no cloud in between; only your own AI is called.
Pricing

Start free

Everything Xervente does is in the free plan. Pro lifts the limits.

Free

$0 forever
  • Up to 3 projects
  • 1 paired phone
  • Autopilot and Review
  • Risky steps and secrets held for you

Pro

Soon priced at launch
  • Unlimited projects
  • Unlimited paired phones
  • Unlimited code slices a day
  • One licence per computer
Private beta

Try Xervente before everyone else

We are opening Xervente to a small group of developers first. Tell us what you use; we will write when your place is ready.

  • Windows and macOS, with the Android and iPhone apps
  • Claude Code first; Cursor, Gemini CLI and Antigravity in beta
  • Your feedback shapes what ships
Questions

Good to know

Does Xervente read or upload my code?

It reads what your agent is doing, on your computer, so it can answer in context. Nothing is uploaded to us. The only thing that leaves your machine is the question it asks your own AI account, exactly like your agent does.

Which agents does it work with?

Claude Code (CLI and the VS Code, Cursor and JetBrains extensions) is fully supported. Cursor, Gemini CLI and Antigravity are in beta. With Windsurf, Xervente records and holds risky steps, but cannot answer the agent, because Windsurf does not allow it.

What if Xervente answers something wrong?

Use Review to check every answer first, or undo a decision in the app and say why: Xervente learns from it and will not make that call again. Risky steps are never decided for you.

Can I turn it off?

Any time. Per project with /xervente off, or everywhere with one switch in the app. Your agent then works exactly as before.

Does my phone need the internet?

Your phone talks straight to your computer on the same network, paired with a QR code. Nothing goes through a Xervente server.